COMPUTER |

GRAPHICS
PM cess ats: Approach |
SECOND EDITION | Caan Essie)

T r y
es © @e@@@
@ @ @ 6080 8 @
@ @ 6.6.0 @
ww ww KA
Ty wom t ow ORE
Brwswt S S
Mw XKAED

OW/
® ~ McGRAW-HILL INTERNATIONAL EDITIONS
. | a] i Computer Science Series
 om Tart
ert - SMATahs
“TDUNDEE UWNIVEeSY
O2 /osf/ot + |-7-95
 COMPUTER GRAPHICS
A PROGRAMMING APPROACH
 COMPUTER
GRAPHICS
A Programming Approach

Second Edition

Steven Harrington
Xerox Corporation

McGraw-Hill Book Company

New York St. Louis San Francisco Auckland Bogota Hamburg
London Madrid Mexico Milan Montreal New Delhi Panama
Paris Sao Paulo Singapore Sydney Tokyo Toronto
vi CONTENTS

An Application 57
Further Reading 60
Exercises 61
Programming Problems = 63

Polygons 70
Introduction 70
Polygons 70
Polygon Representation vA
Entering Polygons G2
An Inside Test 74
Polygon Interfacing Algorithms 76
Filling Polygons 79
Filling with a Pattern 91
Initialization 93
Antialiasing 94
An Application 94
Further Reading 96
Exercises 97
Programming Problems 99

Transformations 107
Introduction 107
Matrices 107
Scaling Transformations 109
Sin and Cos 112
Rotation 113
Homogeneous Coordinates and Translation 116
Coordinate Transformations 118
Rotation about an Arbitrary Point 119
Other Transformations 120
Inverse Transformations 122
Transformation Routines 125
Transformations and Patterns 131
Initialization 136
Display Procedures 136
An Application 139
Further Reading 140
Exercises 141
Programming Problems 143

Segments 146
Introduction 146
The Segment Table 147
Segment Creation 149
Closing a Segment 150
Deleting a Segment 150
 CONTENTS vii

Renaming a Segment 153

Visibility 154
Image Transformation 155
Revising Previous Transformation Routines 157
Saving and Showing Segments 159
Other Display-File Structures 161
Some Raster Techniques 163
An Application 165
Further Reading 166
Exercises 167
Programming Problems 169

Windowing and Clipping 172

Introduction 172
The Viewing Transformation 173
Viewing Transformation Implementation 178
Clipping 181
The Cohen-Sutherland Outcode Algorithm 182
The Sutherland-Hodgman Algorithm 184
The Clipping of Polygons 190
Adding Clipping to the System 195
Generalized Clipping 196
Position Relative to an Arbitrary Line 197
Multiple Windowing 198
An Application 199
Further Reading 200
Exercises 201
Programming Problems 202

Interaction 205
Introduction 205
Hardware 206
Input Device—Handling Algorithms 211
Event Handling 214
Sampled Devices 221
The Detectability Attribute 222
Simulating a Locator with a Pick 225
Simulating a Pick with a Locator 225
Echoing 230
Interactive Techniques 230
Further Reading 239
Exercises 241
Programming Problems 242

Three Dimensions 244

Introduction 244
3D Geometry 244
 BE heise

NET £ yet —

roa <5 at (it a

Pa , ‘chs ‘ehie

Pom hy.2"
ra Sa eeee
dé vess 4
:
" Ages4 Jie, MiAa! %ee
a

Hed :

1 Lei, Coos. ai.

dite
3 Thimeaitag -
LSE INS en G3
ar AB onan ae
Pile ee
 PREFACE

HANDS-ON APPROACH
This book, the second edition of an introductory text on interactive computer graphics,
presents the basic concepts of that field. The approach is somewhat different from that
of previous texts in that it encourages a “hands-on,”’ or “learn-by-doing,’’ attitude.
The book provides guidance for developing a graphics system, suggestions for modifi-
cations and extensions of the system, and application problems which make use of the
system. The advantages of this approach are a high level of student involvement, an
understanding of the systems aspects of graphics software, a realistic feeling for
graphics system capabilities and ease of use, and a greater feeling of accomplishment.

PREREQUISITES
This book is written at a lower level than previous graphics texts. It is intended for stu-
dents with a knowledge of high school algebra, analytic geometry, and trigonometry,
and with at least one high-level language programming course. Although the necessary
information on vectors, matrices, and data structures is presented in the text, students
familiar with these topics will find the material easier to follow. The text should be suit-
able for junior college and undergraduate college levels. Most of computer graphics is
based on the mathematics of analytic geometry; a knowledge of this area is appropriate
and necessary. This book presents a graphics system as a series of programming prob-
lems, so a familiarity with some high-level language is required. The text attempts to
present its algorithms in a language-independent fashion and does not require the
knowledge of any one particular language. It has been used successfully with Pascal,
C, and Fortran.

TIME REQUIREMENTS
The material can be covered at a rate of about one chapter per week. Slightly more
time is needed to cover Chapters 8 to 11. Each chapter also provides a programming as-
signment for a component of the graphics system. The requirement of a weekly pro-

xi
Xii PREFACE

gram is a heavy but not excessive load for students. If necessary, the less instructive
portions of programming projects may be provided by the instructor. At the end of the
course, students have not only a working knowledge of computer graphics fundamen-
tals but also a graphics system, or a collection of graphics programs, of which they can
rightly be proud. There are several suggested programming problems at the end of each
chapter. Some of these problems are much more difficult than others. The difficult
problems are marked with an asterisk; very difficult problems have two asterisks.

ORGANIZATION
The text is organized so that each chapter introduces a new topic in computer graphics.
The progression is such that new chapters build on previous ones, so the order of pres-
entation is rather fixed. The first chapter begins with a very elementary review of ana-
lytic geometry and a discussion of vector generation. The second chapter considers
line- and character-drawing commands and provides the interface between the specific
display devices being used and the rest of the system. It also presents the basic concept
of the display file, or metafile. The third chapter presents polygon surfaces and the ras-
terization problem. The fourth chapter discusses the basic transformations of transla-
tion, scaling, and rotation. The fifth talks about display file segmentation and visibil-
ity. These first five chapters, dealing with the management and interpretation of the
display file, form a natural group when considering intelligent graphics terminals
which maintain their own display files. The sixth chapter completes the discussion of
two-dimensional graphics with a discussion of windowing and clipping. The instructor
may choose to cover Chapter 6 before Chapter 5. This groups the image construction
topics together, followed by the data structuring topic. In the seventh chapter, interac-
tive techniques are discussed. Here again, routines act as an interface between the rest
of the system and the particular graphics input hardware available. In Chapter 8 the
system is generalized to three-dimensional graphics. The topics of three-dimensional
geometry, viewing transformations, parallel projection, perspective projection, and
three-dimensional clipping are discussed. Chapter 9 considers hidden surface and line
removal. It presents two methods in detail: The first is the simple check for removal of
back faces; the second is a priority sort for implementation of the painter’s algorithm.
The tenth chapter considers shading and color. The eleventh and final chapter consid-
ers the drawing of curved lines and surfaces. Arc generations, interpolations, and frac-
tals are covered. h
Throughout the text, the generic masculine pronoun “he” has been used solely
on account of the ease of expression it affords. Such use of the masculine pronoun
should not be interpreted as a wish to exclude women from the use of this text or from
the field in general; in fact, all areas of computer science provide excellent career op-
portunities for women. “The programmer...he” is used only because it is less cumber-
some than “the programmer...he or she.”

RELATION TO STANDARDS
The original edition of this text was modeled after the GSPC CORE system, with
ex-
tensions to raster graphics, hidden surfaces, shading, and curves. Although
the new
 PREFACE Xiilli

edition has modified and extended the system, the basic object-oriented structure and
basic input and output primitives have been preserved. Objects are modeled in world-
coordinate space, and views of the objects are transformed to normalized device coor-
dinates. Individual graphics primitives can be grouped into segments. This approach is
common to many of the current standards, so the student will feel at home with the
GKS, PHIGS, CGI, and CGM standards.

SIMPLICITY VS. EFFICIENCY

There are many places in this book where several different algorithms are possible,
where the problem may be solved many different ways. As a rule, only one solution is
presented in great detail, although others may be described. The particular solution
chosen may not be the most clever or the most efficient. The primary objective was not
to create the best graphics system but to teach the basic graphics principles. The al-
gorithms selected are, therefore, those which were felt to most readily convey these
principles and which were easiest to understand.

APPROACHES TO A GRAPHICS COURSE

There are several ways of using this book. One method is to have students implement
the described graphics system. With this approach, the student gets a good understand-
ing of how the components of a graphics system work and interact. Since each exten-
sion builds on the previous week’s work, the instructor may need to provide solutions
to previous assignments. One danger in providing algorithms as complete as those
given here is a temptation to copy them without really understanding them. This
should be countered with classroom explanation and by assigning some of the exer-
cises which are included.
A second approach is to use the presented graphics system as a base for exten-
sions and modifications. Alternative algorithms and methods may be explored. The
book provides a well-documented basic support system, or test bed, on which these
methods can be actually programmed. The actual implementation and use of an ap-
proach can provide insights into the details of its behavior and its system implications,
which may be missed when the algorithm is discussed in isolation. Programming prob-
lems suggesting extensions have been included.
A third approach is to emphasize using a graphics system rather than building
one. With this approach, the algorithms serve as examples of graphics system routines,
but not as implementation guides. Students should have this system or some other
equivalent system available. Classroom lectures can still explain graphics systems and
how they work, but programming projects use these features rather than build them.
Applications-oriented programming problems are provided for this purpose.

CHANGES IN THE SECOND EDITION

Computer graphics is a rapidly developing field, and even though this is an introduc-
tory text, some revision is needed in order to stay current. In addition, weaknesses in
the original text have been addressed. Actual changes include a discussion of the popu-
XIV PREFACE

lar Bresenham vector generation algorithm; an introduction to antialiasing techniques

and imaging with pixel arrays and patterns; a discussion of the Cohen-Sutherland Out-
code clipping algorithm and use of the Blinn shading model; discussions of ray trac-
ing, halftones, and several color topics; and an introduction to Bezier curves, surface
patches, and fractals. Three-dimensional clipping was included with the other three-di-
mensional extensions in Chapter 8, instead of occupying its own chapter. The hidden
surface and line chapter was completely redone, providing much more discussion of al-
ternate approaches and less on the algorithm details. It follows Fuch’s approach to sort-
ing polygons, simplifying comparison techniques, and does not attempt to discuss
polygon decomposition. Other changes include a greatly expanded set of references,
an increased selection of exercises and programming problems, and an eight-page in-
sert of full-color photographs (Plates 1 to 16) in Chapter 10.

FURTHER SUPPORT
One thing learned through the first edition of the book is that the text creates the need
for information about implementations of the graphics system described. What im-
plementations are available? What languages and computing environments have been
tried? What problems or bugs were discovered? What extensions have been carried
out? With this edition I shall attempt to act as a clearinghouse for this information. As
such, I shall welcome hearing of implementations and requests for. implementations.
All inquiries can be sent to Steven Harrington, Xerox Corporation, W128-29E, 800
Phillips Rd., Webster, NY 14580.

APPRECIATION
I would like to express my appreciation and thanks to those who reviewed this edition
of the text: Steve Cunningham, Rollin Dix, William Grosky, David McAllister, and
Spencer Thomas. Their comments and suggestions were very valuable, and I am most
grateful for their assistance.
Steven Harrington
 CHAPTER

GEOMETRY
AND LINE
GENERATION

INTRODUCTION
Perhaps our age will be known as the Information Revolution or the Computer Revolu-
tion, for we are witnessing a remarkable growth and development of computer technol-
ogy and applications. The computer is an information processing machine, a tool for
storing, manipulating, and correlating data. We are able to generate or collect and pro-
cess information on a scope never before possible. This information can help us make
decisions, understand our world, and control its operation. But as the volume of infor-
mation increases, a problem arises. How can this information be efficiently and effec-
tively transferred between machine and human? The machine can easily generate ta-
bles-of numbers hundreds of pages long. But such a printout may be worthless if the
human reader does not have the time to understand it. Computer graphics strikes di-
rectly at this problem. It is a study of techniques to improve communication between
human and machine. A graph may replace that huge table of numbers and allow the
reader to note the relevant patterns and characteristics at a glance. Giving the computer
the ability to express its data in pictorial form can greatly increase its ability to provide
information to the human user. This is a passive form of graphics, but communication
can also be a two-way process. It may be convenient and appropriate to input graphical
information to the computer. Thus there are both graphical input and graphical output
devices. It is often desirable to have the input from the user alter the output presented
by the machine. A dialogue can be established through the graphics medium. This is
termed interactive computer graphics because the user interacts with the machine. Com-
puter graphics allows communication through pictures, charts, and diagrams. It offers
2 CHAPTER ONE

a vital alternative to the typewriter’s string of symbols. The old adage ‘‘A picture is
worth a thousand words’’ is certainly true. Through computer graphics we can pilot
spaceships; walk through buildings which have yet to be byilt; and watch bridges col-
lapse, stars being born, and the rotations of atoms. There are many applications for
computer graphics. Management information may be displayed as charts and dia-
grams. Scientific theories and models may be described in pictorial form. (See Plates 1
through 4.) In computer-aided design we can display an aircraft wing, a highway lay-
out, a printed circuit board, a building ‘‘blueprint,’’ or a machine part. (See Plate 15.)
Maps can be created for all kinds of geographic information. Diagrams and simula-
tions can enrich classroom instruction. The computer has become a new tool for the
artist and animator. (See Plates 5 through 14.) And in video games, computer graphics
provides a new form of entertainment.
Over the years many graphics display devices have been developed. There are
also many software packages and graphics languages. The problem with such diversity
is that it makes it difficult to transfer a graphics program from one installation to
another. In the late 1970s, the Graphics Standards Planning Committee of the Associa-
tion for Computing Machinery developed a proposal for a standard graphics system
called the CORE system. This system provided a standardized set of commands to con-
trol the construction and display of graphic images. The commands were independent
of the device used to create or to display the image and independent of the language in
which the graphics program was written. The CORE system defined basic graphics
primitives from which more complex or special-purpose graphics routines could be
built. The idea was that a program written for the CORE system could be run on any
installation using that system. The CORE system contained mechanisms for describing
and displaying both two-dimensional and three-dimensional structures. However, it
was developed just before the reduction in the cost of computer memory made possible
economical raster displays (which allow solid and colored areas to be drawn). It there-
fore lacked the primitives for describing areas and could only create line drawings. Ex-
tensions were soon proposed to provide the CORE system with raster imaging primi-
tives.
A second standard called the graphics kernel system (GKS) was developed in
Europe, and it has been steadily gaining in popularity. The GKS system was heavily in-
fluenced by CORE, and the minimal GKS implementation is essentially identical to
CORE’s two-dimensional subset. The GKS standard did contain primitives for imag-
ing areas and colors, but it did not contain the constructs for three-dimensional objects.
It introduced the concept of a workstation, which allowed a single graphics program to
control several graphics terminals.
Another graphics standard is the programmer’s hierarchical interactive graphics
standard (PHIGS). It takes input and output functions and viewing model from CORE
and GKS, but it is a programmer’s toolbox, intended for programming graphics appli-
cations. It contains enhanced graphics program structuring features. Two other
graphics standards are the computer graphics metafile (CGM) and the computer
graphics interface (CGI). The CGM isa file format for picture information that allows
device-independent capture, storage, and transfer. The CGI is a companion standard
which provides a procedural interface for the CGM primitives.
 GEOMETRY AND LINEGENERATION 3

In this book we are going to present the algorithms for constructing a graphics
system which have the flavor of the CORE and GKS standards and, in some areas, go
beyond them.
We begin our discussion of computer graphics with the fundamental question of
how to locate and display points and line segments. There are several hardware devices
(graphics terminals) which may be used to display computer-generated images. Some
of these will be discussed in Chapter 2. Before we talk about the devices which display
points, we shall review the basic geometry which underlies all of our techniques. We
shall consider what points and lines are and how we can specify and manipulate them.
We conclude this chapter with a discussion of how the mathematical description of
these fundamental geometric building blocks can be implemented on an actual display
device. Algorithms are presented for carrying out such an implementation for a line
printer or common cathode ray tube (CRT) display. These algorithms will allow us (if
needed) to use the line printer or CRT as a somewhat crude, but effective, graphics dis-
play device for demonstrating the graphics principles described in the rest of the text.

LINES
We can specify a point (a position in a plane) with an ordered pair of numbers (x, y),
where x is the horizontal distance from the origin and y is the vertical distance. Two
points will specify a line. Lines are described by equations such that if a point (x, y)
satisfies the equations, then the point is on the line. If the two points used to specify a
line are (x;, y,) and (Xx, y2), then an equation for the line is given by
YOM 2M YO, DAWES (11)
Ke a XT Xp — X

This says that the slope between any point on the line and (x,, y,) is the same as the
slope between (X2, yz) and (x, yj).
There are many equivalent forms for this equation. Multiplying by the de-
nominators gives the form

# (x —X,) (Y2-y1) = Y-Yi) (2 -%)) (1.2)

A little more algebra solving for y gives

Noises Coraees Co hada (1.3)

Xo — X}

or
y= mx + b wg (1.4)
where | choc 04
Mee - \ Ly
) Vs ‘ KY VL

: Decst pn, Ne
Xo — Xy s

an 2 2? { yy isi 4h

b= y; — mx; \ N oy
rAd \
 4 CHAPTER ONE

This is called the slope-intercept form of the line. The slope m is the change in
height divided by the change in width for two points on the line (the rise over the run).
The intercept b is the height at which the line crosses the y.axis. This can be seen by
noting that the point (0, b) satisfies the equation of the line.
A different form of the line equation, called the general form, may be found by
multiplying out the factors in Equation 1.2 and collecting them on one side of the equal
sign.

(Y2-Y1) X- (KQ-XDY + yi -Xiy2 = 0 (1.5)

or
x +sy+t=0 (1.6)
where possible values for r, s, and t are

P= (Yo
—Va)
S = = Gy x)

OS XoVe a a1 Ya
We say possible values because we see that multiplying r, s, and t by any common fac-
tor will produce a new set of r’, s’, and t’ values which will still satisfy Equation 1.6
and, therefore, also describe the same line. The values for r, s, and t are sometimes
chosen so that

cts =] (1.7)
Comparing Equations 1.4 and 1.6 we see that
r
m= ——
S
and (1.8)
t
b= - —
S
Can we determine where two lines will cross? Yes, it is fairly easy to
determine
where two lines will cross. By the two lines crossing we mean that
they share some
point in common. That point will satisfy both of the equations for
the two lines. The
problem is to find this point. Suppose we give the equations for
the two lines in their
slope-intercept form:

line 1: yi =m, 1 x +b 1
(1.9)
line 2: y = mx + by
Now if there is some point (x;, y;) shared by both lines, then

Yi; = Migxy Hsby and Yi = Mx; + by (1.10)

will both be true. Equating over yj; gives

MX; ot b, —— MX; at bz (1.11)

 GEOMETRY AND LINE GENERATION 5

Solving for x; yields

sab som B
cen
ek (1.12)
m, i Mp»

Substituting this into the equation for either line 1 or line 2 gives

b.m, — b,;m
a eee (1.13)
m, a My

Therefore, the point

(b, — b,; bom, — re) (1.14)

’
m, — M) m, — mM)

is the intersection point. Note that two parallel lines will have the same slope. Since
such lines will not intersect, it is not at all surprising that the above expression results
in a division by zero. When no point exists, we cannot solve for it.
If the Equation 1.6 form is used to describe the lines, then similar algebra yields
an intersection point which is given by

os sae Sot) tiI> ary 2 (1 15)

Sof; — S$jT2 Sol) — S)fp

LINE SEGMENTS
What are line segments? Our equations for lines specify all points in a given direction.
The lines extend forever both forward and backward. This is not exactly what we need
for graphics. We would like to display only pieces of lines. Let’s consider only those
points on a line which lie between two endpoints p, and p>. (See Figure 1-1.)
This is called a line segment. A line segment may be specified by its two end-
points. From these endpoints we can determine the equation of the line. From this
equation and the endpoints we can decide if any point is or is not on the segment. If the
endpoints are p; = (X;, y;) and py = (X», yz) and these yield equation y = mx + b
(or rx + sy + t = 0), then another point p3; = (x3, y3) lies on the segment if

1. y3 = mx3 + b(orrx3 + sy3 + t = 0) @z

2. min(xX,, X2) S x3 S max(x, X2)
3. min(y;, y2) = y3 = max(yj, y2)

P2

FIGURE 1-1
Py A line segment.
6 CHAPTER ONE

Here the notation min(x,, xX) means*the smallest of x, and x, and max(x;, Xz) means
the largest of the two numbers.
There is one more useful form of the line equation called the parametric form be-
cause the x and y values on the line are given in terms of a parameter u. This form is
convenient when considering line segments because we can construct it such that line-
segment points correspond to values of the parameter between 0 and 1. Suppose we
want the line segment between (x,, y;,) and (x2, y2). We wish the x coordinate to go
uniformly from x, to xz. This may be expressed by the equation
X = xX, + (X)-X,)u (1.16)

When uis 0, x is x. As u increases to 1, x moves uniformly to x,. But for a line seg-
ment, we must have the y coordinate moving from y, to y, at the same time as x
changes.

y =y, + 2-y,)u (1.17)

The two equations together describe a straight line. This can be shown by equating
over u. A little algebra recovers the line equation. Note that with this form, we can gen-
erate the point on the line segment by letting u sweep from 0 to 1; also, given the
parameter value for a point on the line, we can easily test to see if the point lies within
the segment boundaries.
How long is a line segment? If we are given the two endpoints of a line segment
Pp, and p, we can determine its length L. Construct a right triangle p,p>A by attaching a
vertical line to pz and a horizontal line to p,. (See Figure 1-2.)
The Pythagorean theorem states that the square of the length of the hypotenuse
(P,P) is equal to the sum of the squares of the lengths of the two adjacent sides (p,A
and p2A). If we call the coordinates of p; (x;, y,), and the coordinates of p> (X, y>),
then A will have coordinates (x, y,), and the length of the segment L will be given by

L? = (&y— x) (¥2— yy)? (1.18)

so

L = [(%) — x1)? + (yo — y,)"]!” (1.19)

Yar DP?

A =lh

ah ran
1
1X2
_—- FIGURE 1-2
Line segment length.
 GEOMETRY AND LINE GENERATION 7

What is the midpoint of a line segment? The midpoint ofa line segment is often
useful and easy to calculate. The point halfway between the endpoints of a segment
will have an x coordinate halfway between the x coordinates of the endpoints, and a y
coordinate halfway between the y coordinates of the endpoints. (See Figure 1-3.)
Therefore, the midpoint is

G Ve |: +X) (¥; + Yo)

m? Ym A - * ’ 2 (1.20)

PERPENDICULAR LINES
Can we tell if two lines are perpendicular? We can determine if two lines are perpen-
dicular by examining their slopes. Suppose we have two lines
y=m,x+b,

and C1221)
y = mx + b,
If the first line is perpendicular to the second, then a line parallel to the first (that is, a
line with the same slope) will also be perpendicular to the second. For example, y =
m,x should be perpendicular to y = mx + by. The same argument applies to the sec-
ond line: y = m x will be perpendicular to y = m,x. Now these are two lines which
intersect at the origin. (See Figure 1-4.)
Consider a point (x,, y,) on the line y = m,x so that y; = m,x, and a point
(X>, Y2) ON y = Mpx SO that yp = MpX>. The three points (x, y,), (Xz, yz), and (0, 0)
form a triangle. If the two lines are perpendicular, they will form a right triangle and
the Pythagorean theorem will apply. The distance between (0, 0) and (x,, y,) squared
plus the distance between (0, 0) and (x2, yz) squared will equal the square of the
hypotenuse between (x, y;) and (X,, yz). The distance formula gives

FIGURE 1-3
The midpoint of a line segment.
8 CHAPTER ONE

FIGURE 1-4
Construction for test for perpendicu-
lar lines.

xi + yj + x3 + yo = (% — %2) + (1 — Ya) (1.22)

Simplifying gives
O = —2y1y2 — 2X1X2
or (1.23)
Viwe™ XQ
Ry RET
but since y,; = m,x, and yp = M Xp, we have

i= eh (1.24)
M3
Therefore, if two lines are perpendicular, the slope of one will be the negative recip-
rocal of the slope of the other.
From Equations 1.8 and 1.22 we see that two lines ee in the general form
for a line are perpendicular if
T,I> + $);So = 0 (1.25)

It also follows that

(Xp) — Xa1)(Xp2 — Xa2) + (Voi — Yai) (¥b2 —Ya2) = 0 (1.26)

for (Xai, Yai) and (Xp), pi) On line 1, (xa2, Yao) and (Xp2, Yp2) On line 2.

DISTANCE BETWEEN A POINT AND A LINE

We shall derive one more formula in our review of geometry. What is the distance be-
tween a point and a line in a plane? Suppose we have a point (Xo, Yo) and a line rx +
 GEOMETRY AND LINE GENERATION 9

sy + t = 0, where r, s, and t were chosen to satisfy Equation 1.7. We can find a line
which is perpendicular to the given line and contains the given point. It is

—sx + ry + (Sxg + ryo) = 0 (127)

We can determine the intersection point of the original line and this perpendicular. (See
Figure 1-5.) Using Expression 1.15 we find it to be

(s (sx — rYo) — rt ,-st — r (8Xp — r¥0)) (1.28)

Now we can use the distance formula (Equation 1.19) to determine the distance be-
tween the point (Xo, Yo) and this intersection point. This is what we mean by the dis-
tance between the point and the line.

L = ({Xo — [s (8X9 — r¥o) — rt]}? + {yo — [—st — 1 (sxp — ryo)]}*)' (1.29)

This will reduce down to

L = |rxp + syo + t | (1.30)

Notice that this is just the magnitude of the value obtained by substituting the coordi-
nates of the point into the expression for the line. When the expression is zero, the
point is on the line, while other values give the distance of the point from the line. Re-
member that this only works because we have chosen values for r, s, and t which

(Xo; Yo)

Distance

(s(SXq — r¥o) —rt, —st — r(sxp — ryo))

rmx+sy+t=0

FIGURE 1-5
Distance between a point andaline.
 10. CHAPTER ONE

satisfy Equation 1.7, and in fact this simple distance relationship is the motivation for
this choice.

VECTORS
What is a vector? A vector has a single direction and a length. A vector may be denoted
[D,, D,], where D, indicates how far to move along the x-axis direction and D, indi-
cates how far to move along the y-axis direction. (See Figure 1-6.)
Unlike line segments, vectors have no fixed position in space. They tell us how
far and what direction to move, but they do not tell us where to start. The idea of a vec-
tor is useful because it closely parallels the manner in which a pen draws lines on paper
or an electron beam draws lines on a cathode ray tube. The command to the pen may
be to move so far from its current position in a given direction.
Two vectors may be added by adding their respective components.

Vi + V2 = (Dy, Dyl + Dior Di) = (Dari Do, Dy + Det Ga

We can picture this in terms of pen movements. Suppose we start at some point A. The
first vector moves the pen from point A to point B; the second, from point B to point
C. The right-hand side of the above equation produces a single vector which will move
the pen directly from point A to point C.
We can also multiply a vector by a number by multiplying each of its compo-
nents. :

nV = n[D,, Dy) = (aD, nD] (1.32)

This preserves the vector’s direction but changes its magnitude. A measure of thatmag-
nitude is given by the vector’s length.

|Volpe De aE ee (1.33)
If we multiply a vector by the reciprocal of its length, the result is a vector with length
equal to 1. Such vectors are called unit vectors. They conveniently capture the direc-
tion information.
NER.

<

FIGURE 1-6
-———— D, ——_____ A vector.
 GEOMETRY AND LINEGENERATION I]

We sometimes use vectors as a shorthand way of expressing operations on all

coordinates. For example, the parametric equations for a line (Equations 1.16 and 1.17)
can be combined into the vector form

Vie, omy,Vv (1.34)

whereV = [x, y], V; = [x,, y,], and V, = [X, yp].
We shall consider additional vector operations in later chapters as the need
arises.
Wh a I vat a

PIXELS AND FRAME BUFFERS )

How does all this apply to an actual graphics display? To begin with, the mathematical
notion of an infinite number of infinitesimal points does not carry over to the actual
display. We cannot represent an infinite number of points on a computer, just as we
cannot represent an infinite quantity of numbers. The machine is finite, and we are lim-
ited to a finite number of points making up each line (usually no more than a few
hundred to a few thousand). Themaximum number of distinguishable points whicha
line may have is a measure of the resolution of the display device. The greater the
number of points, the higher the resolution. This limitation in the number of points
may not bother us too much, because the human eye does not notice much detail finer
than 1000 points per line segment. Since we must build our lines from a finite number
of points,each point must have some size and so is not really a point at all. Itis called
a1pixel (short for picture element). The pixel is the smallest addressable screen element.
‘Ttisthe smallest piece of the display screen which we can control. Each pixelhas a
name or address. The names which identify pixels correspond to the coordinates which
identify points. Computer graphics images are made by setting the intensity and color
of the pixels which compose the screen. We draw line segments by setting the inten-
sities, that is, the brightness, of a string of pixels between astarting pixel and an end-
ing pixel. We can think of the display screen as a grid, or array, of pixels. We shall give
integer coordinate values to each pixel. Starting at the left with 1, we shall number
each‘column. Starting at the bottom with 1, we shall number each row. The coordinate
(i, j) will then give the column and row of a pixel. Each pixel will be centered at its
coordinates. (See Figure 1-7.)
We may wish to place the intensity values for all pixels into an array in our com-
puter’s memory. Our graphics display device can then access this array to determine
the intensity at which each pixel should be displayed. This array, which contains an in-
ternal representation of the image, is called the frame bufer. It collects and stores pixel
values for use by the display device. _ i;

VECTOR GENERATION
The process of ‘‘turning on’’ the pixels for a line segment is called vector generation. If
we know the endpoints which specify the segment, how do we decide which pixels
12s CHAPTER ONE

FoRa asdl cets

oer eels ana
cf [OL aTaea a
os a ee
fr fs ie Ba
{sakes ane
Pisceif eee 5eli
a
ee eee
Seles an? alee a Seto A pixel.

should have their intensity changed? There are several approaches to this problem. We
shall present two examples here. The first is a general algorithm, while the second is a
more efficient version for the case where the line segment endpoints are integers.
The problem is to select pixels which lie near to the line segment. We might try
to turn on every pixel through which the line segment passes, but there is a problem
with this approach. It would not beeasy to findall such pixels,and since vector gener-
ation may be performed often, especially for animated displays or complex images, we
want it to be efficient. Another problem is that the apparent thickness of the line would
change with slope and position. An alternative would be to step along the columns of
pixels, and for each column ask which row is closest to the line. We could then turn on
the pixel in that row and column. We know how to find the row because we can place
the x value corresponding to the column into the line equation, solve for y, and note
which row it is in. This will work for lines with slopes between —1 and 1 (lines which
are closer to being horizontal than vertical). But for the steeply rising or falling linés,
the method will leave gaps. This failing can be overcome if we divide the lines into
two classes. For the gentle slopes (-1 < m < 1) there are more columns than rows.
These are the line segments where the length of the x component D, = ( x, — x, ) is
longer than the length of the y component Dy = (Yb — Ya.» that is, ag For
these cases, we step across the columns and oe for the rows.‘ Forthe sharp slopes
where |D,| < , we Step up the rows and solve for the columns.
It would still be very inefficient if we actually had to solve the line equation to
determine every pixel, but fortunately we can avoid this by taking uniform steps and
using what we learn about the position of the line at each column (or row) to determine
its position at the next column (or row). Consider the gentle slope case where we step
across the columns. Each time we move from one column to the next, the value of x
changes by |. But if x always changes by exactly 1, then y will always change by
exactly m (the slope). The change iny is

127 Yi+1.— ¥i =" Cx, | +,b) = (mx; + 5)

ar| | (1.35)
= m(x\4,—x;) = m(1) = m
 GEOMETRY AND LINEGENERATION 13

This means that as we step along the columns, we can find the new value
of y for the
line by just adding m to the y position at the previous column. So our
approach so far
is first to find the column at the left end-of the line segment and the y value
at that col-
umn, and then to step through the columns which the line segment crosses,
adding m
to the y value each time to get a new y value. The y values will be used
to select a
row at each column, and we turn on a pixel at the selected row and column.
(See
Figure 1-8.)
The vector generation algorithms (and curve generation algorithms) which step
along the line (or curve) to determine the pixels which should be turned on are some-
times called digital differential analyzers (DDAs). The name comes from the fact that we
use the same technique as a numerical method for solving differential equations. For a
line segment, we solve the differential equation for a straight line and plot the result in-
stead of printing it.
We shall make one further refinement in our approach. Instead of determining
the full y value of the line segment for each column, we shall only keep track of the
current height above the closest row boundary. At each step, the height increases by m
along with y. We can check the height at each step to see if we have moved into a new
row. If we have entered a new row, then we change the row value used to select pixels
and also subtract 1 from our height so that we will now be measuring height from the
new row boundary. For lines with negative slopes, we could let the height decrease and
check for crossing of a lower row boundary, but instead we use the absolute value of

FIGURE 1-8
Turn on pixels with centers closest to the line segment.
 14 CHAPTER ONE

height and slope and check as it steps up. This is in order to be consistent with the al-
gorithm presented in the next section.
For lines with sharp slopes, a similar procedure is used, only the roles of x and y
are exchanged and a new ‘‘height’’ isfound for a new row by adding 1/m to the height
value for the old row.
Now a few more words on how our particular vector generator finds the starting
point. For the gentle slope cases, we shall turn on pixels in the columns that have a
center line which crosses the line segment. We shall center the columns and rows on in-
teger coordinate values, so if x, is the left end of the line segment and x, is the right
end, then columns between c = CEILING(x,) and f = FLOOR(x,) inclusive are af-
fected. (The function CEILING returns the smallest integer which is greater than or
equal to its argument, and FLOOR returns the largest integer which is less than or
equal to its argument.) Our starting y position will correspond to the point at x = c,
and might not be the endpoint y,. The starting y value may be determined from the line
equation as follows: Gy wi
% \ wif sect ages ,
dee bie cesar d = mc + b = mc + (y,—mx,) (1.36)
Wyrw—rere 5 wr a ead
hey Wwe CO o>) = ya + m(c - x,) a
> To find the index of the closest row, we round the value of y. Rounding may be done by
2 adding 0.5 and then using the FLOOR function.
Poaqnea iv r = FLOOR(y + 0.5) (1.37)
14-05-5944 : or “= |

FIGURE 1-9
Height of the line above pixel-row boundaries.
/ re
A ~9
\Y |py. VO OXON WS XW 3 VIN \n CNA vy ‘

e ee
xXLe - 14
a GEOMETRY AND LINE GENERATION 15
So if =

To find the height of y above the row boundary, we take the difference of y (where the
line is) and r (the row index). If we then add m to this, we have the height value which
should be checked for the next column and so on. (See Figure 1-9.)
We shall now present the full algorithm.

1.1 Algorithm VECGEN(XA, YA, XB, YB, INTENSITY) For changing pixel values
of the frame buffer along a line segment 5
Arguments XA and YA are the coordinates of one endpoint
XB and YB are the coordinates of the other endpoint
INTENSITY is the intensity setting to be used for the vector
Global FRAME the two-dimensional frame buffer array
Local DX, DY the vector to be drawn
R and C the row and column indices for the pixel to be changed
F the stopping index
D the line segment coordinate at the starting point
H the difference between the line segment and the row index
M the slope of the line segment
MI the change in H when a boundary is crossed
BEGIN |
determine the components of the vector Vi
DX < XB - XA;
DY <— YB - YA;
decide on whether to step across columns or up rows
- IF|DX| > |DY| THEN
BEGIN
the gentle slope case =
M < DY/ Dx; aod ie <<a
set up of starting point depends on which point is leftmost
IF DX > 0 THEN
BEGIN
C < CEILING(XA);
D<YA + M « (C- XA);
F — FLOOR(XB);
END
ELSE
BEGIN
C <— CEILING(XB);
D<YB + M « (C - XB);
F — FLOOR(XA);
END;
R < FLOOR(D + 0.5);
H<R-D+M;
IF M > 0 THEN
BEGIN
the positive slope case
MI <—M-1;
now step through the columns
WHILE C = FDO
 16 CHAPTER ONE

BEGIN
set the nearest pixel in the frame buffer
FRAME[C, R] <— INTENSITY;
next column
CeCe;
should row change .
IF H = 0.5 THEN ve
BEGIN
Reheat fs
H<—H + M1;
END
ELSE H <—H + M;
END;
END
ELSE
BEGIN
then negative slope case
M <-—M;
H <—-H;
M1 <M-1;
WHILE C = FDO
BEGIN
set the nearest pixel in the frame buffer
FRAME[C, R] < INTENSITY:
next column
C<-CHl;
should row change
IF H > 0.5 THEN
BEGIN
R<R-1;
H<H + M1;
END
ELSEH <—H + M,;
END;
END;
END
ELSE
BEGIN
Ns
the sharp slope case
‘
IF DY = 0.0 THEN RETURN:
here the above steps are repeated
with the roles of x and y interchanged
END;
RETURN;
END;

The actual algorithm contains a coupl

e of things which were not brought
the above discussion. The determinat up in
ion of the Starting values is complica
ted by not
 GEOMETRY AND LINE GENERATION 17

knowing a priori whether XA is left or right of XB, and YA above or below YB. The al-
gorithm also contains a test on the sign of the slope. It is used to select a loop which
moves the line in the proper direction.

BRESENHAM’S ALGORITHM
The above algorithm was chosen because it can be revised into a very efficient and
popular form known as Bresenham’s algorithm for the special case where the coordi-
nates of the line segment endpoints are integers. The attractiveness of Bresenham’s
algorithm is that it can be implemented entirely with integer arithmetic. Integer arith-
metic is usually much faster than floating-point arithmetic. Furthermore, Bresenham’s
algorithm does not require any multiplication or division. To derive the algorithm, first
consider the effect of integer coordinates on determining the starting point. The start-
ing point will just be the endpoint of the line segment. No calculation is needed to
move along the line to a pixel center, because it is already there. This eliminates one
place where the line slope was needed, but we also used the slope to update H. How
can we revise our test for new rows so that it requires only integer arithmetic? To
simplify the discussion, consider just the gentle slope case. The test was
H>0.5 (1.38)
where
H<-H+M

or
H<-H+M-1
at each column. Note first that we can rewrite the test as
H-0.5>0
Now we can multiply by 2 to get
2H-1>0 (1.39)
r

Notice how the 0.5 fraction is removed. But H still has a fractional part, which arises
from the denominator in M that is added in at each step. To remove it, we multiply by
DX. Assuming we have arranged the endpoints so that DX is positive, the test is then
2DxX H- DX > 0
Suppose we define G as
G = 2DX H- DX (1.40)

The test is simply

G>0 (1.41)

Then how does G change from one column to the next? Solving for H gives
18 CHAPTER ONE

G + DX
H = ——— (1.42)
2DX
If a .
Hood “a ode
then
Gia tes Oa oa DY
iDD auean g ‘DX
or
Gewese, tee (1.43)

For the case

Bee Moja cee Mah
we get
Grew Gog + 2DY — 2DX (1.44)
Calculating G requires only additions and subtractions and can be done entirely with
integers. The trick then is to use the test G > 0 to determine when a row boundary is
crossed by the line instead of the test H > 0.5. For integer endpoints, the initial value
of H is M, so the initial value of G is
DY
G = 2DX (me = DX) = 2DY = Dx (1.45)

For each column, we check G. If it is positive, we move to the next row and add
2DY — 2DX to G. Otherwise, we keep the same row and add 2DY to G. The full al-
gorithm follows.

1.2 Algorithm BRESENHAM(XA, YA, XB, YB, INTENSITY) For changing pixel
values of the frame buffer along a line segment with integer endpoints
Arguments XA and YA are the coordinates of one endpoint
XB and YB are the coordinates of the other endpoint
INTENSITY is the intensity setting to be used for the vector
Global FRAME the two-dimensional frame buffer array \ .
Local DX, DY the vector to be drawn s
R, C the row and column indices for the pixel
F the final row or column
G for testing for a new row or column
INC1 increment for G when row or column is unchanged
INC2 increment for G when row or column changes
POS-SLOPE aflag to indicate if the slope is positive
BEGIN
determine the components of the vector
DX <— XB - XA;
DY < YB - YA;
 GEOMETRY AND LINE GENERATION 19

determine the sign of the slope

POS-SLOPE < (DX > 0);
IF DY < 0 THEN POS-SLOPE < NOT POS-SLOPE;
decide on whether to step across columns or up rows
IF |DX| > |DY| THEN
BEGIN
this is the gentle slope case
IF DX > 0 THEN
BEGIN
C<XA;
R<—YA;
F <— XB;
END
ELSE
BEGIN
C:<—— XB:
R = YB;
F— XA;
END;
INC1 <2 « |DY|;
G <2 * |DY| - |DX];
INC2 < 2 « ({DY| - |DX));
IF POS-SLOPE THEN
BEGIN
now step across line segment
WHILE C = FDO
BEGIN
set nearest pixel in the frame buffer
FRAME[C, R] < INTENSITY;
next column
C<C+l;
should row change
IF G = 0 THEN
BEGIN
R<R+1;
G<G + INC2;
END
ELSE G <G + INCI,
END;
END
ELSE
BEGIN
WHILE C = FDO
BEGIN
set nearest pixel in the frame buffer
FRAME[C, R] < INTENSITY;
next column
C<CH 1;
should row change
20 CHAPTER ONE

IF G > 0. THEN
BEGIN
R<R-l1;
G< INC2; :
END
ELSE G —G + INCI;
END;
END;
END
ELSE
BEGIN
this is the sharp slope case
here the above steps are repeated
with the roles of x and y interchanged
END;
RETURN;
END;

We have seen how the BRESENHAM algorithm is more efficient than the VEC-
GEN algorithm in that it requires only integer addition and subtraction. We might ask,
why not always use it? Why not always round line segment endpoints to integers be-
fore vector generation? In fact, many systems will do just that, but note that in round-
ing the line segment endpoint positions, errors are introduced. These errors can be seen
in cases where two overlapping line segments are drawn on the display (as when two
objects are shown side by side). Another example occurs when a new line segment
with a different INTENSITY overlaps an old line segment; we might like to change the
old segment’s pixels to the appropriate values for the new segment. If we deal with
exactly the sarne line equation for both line segments, then the same pixels should be
selected; but if errors in the endpoint positions are introduced, the line equations might
not match, and the new line segment can leave pixels from the old line segment peek-
ing out from behind it. Note, however, that even our VECGEN algorithm can have er-
rors introduced in the endpoint position as the result of round-off in the floating point
arithmetic of the machine.
Algorithms for vector generation (such as the ones we’ve seen) may be im-
plemented in hardware where speed is important. This is usually done for displays
which avoid the cost of a large frame buffer memory by letting the vector generator di-
rectly control the drawing instrument (usually an electron beam in’a cathode ray tube).
Instead of setting the (x, y) element in a frame buffer, the pen (or electron beam) is
moved to position (x, y), where it illuminates the pixel on the screen.
|

ANTIALIASING OF LINES
Many displays allow only two pixel states, on or off. For these displays, lines may
have a jagged or stair-step appearance when they step from one row or column to the
next. The lower the resolution, the more apparent the effect. This is one aspect of a
 GEOMETRY AND LINEGENERATION 21

phenomenon called aliasing. Aliasing produces the defects which occur when the scene
being displayed changes faster or more smoothly than every two pixels. Displays
which allow setting pixels to gray levels between black and white provide a means to
reduce this effect. The technique is called antialiasing, and it uses the gray levels to
gradually turn off the pixels in one row as it gradually turns on the pixels in the next.
(See Figure 1-10.)
The vector generation algorithms can be modified to perform antialiasing. Re-
member that for gentle sloped lines, we in effect examined the line position for each
column index and decided which row was closest. The line segment would lie between
two pixels, and we picked one. Suppose that instead of picking the closest, we turned
them both on. We should choose the intensity values according to a function of the dis-
tance between the pixel index and the line segment so that the pixel closest to the line
receives most of its intensity. The sum of the intensity values for the two pixels should
match the total intensity value for the line. The function used can be a simple or a com-
plex expression based on intensity patterns, pixel shapes, and how lines cover them. In
general, we want the pixel’s intensity to match the amount of the line which covers its
area. Antialiasing with complicated functions can still be done efficiently by storing
the function values in a table. The table is then used to look up the intensity for a dis-
tance between the pixel index and the line. (See Figure 1-11.)

THICK LINE SEGMENTS

Raster displays allow the display of lines with thickness greater than one pixel. To pro-
duce a thick line segment, we can run two vector generation algorithms in parallel to
find the pixels along the line edges. As we step along the line finding successive edge
pixels, we must also turn on all pixels which lie between the boundaries. For a gentle
sloping line between (x,, ya) and (Xp, yp) with thickness w, we would have a top
boundary between the points (x,, y, + Wy) and (Xp, Y» + Wy) and a lower boundary
between (x,, Y, — Wy) and (X,, yp — Wy) where wy is given by
2 24 1/2
cn (w re 1) [(Xp Xa) Ai (Yo Ya) ] (1.46)

r ; 2 Xp EeX,|

This is the amount by which the boundary lines are moved from the line center. The
(w — 1) factor is the desired width minus the one-pixel thickness we automatically re-
ceive from drawing the boundary. We divide this by 2 because half the thickness will
be used to offset the top boundary, and the other half to move the bottom boundary.
The factor containing the x and y values is needed to find the amount to shift up and

oe ec eee e©0220@808060800
sss eee eoeceSOSSSSOGeeeeeoerrr+::

FIGURE 1-10
Antialiasing of a line.
22 CHAPTER ONE

iatet

i ite >
FIGURE 1-11
Using vertical distance from the pixel to determine intensity.

down in order to achieve the proper width w as measured perpendicular to the line di-
rection, not up and down. (See Figure 1-12.) Sharply sloping lines can be handled
similarly with the x and y roles reversed.

CHARACTER GENERATION
Along with lines and points, strings of characters are often displayed to label and anno-
tate drawings and to give instructions and information to the user. Characters are al-
most always built into the graphics display device, usually as hardware but sometimes
through software. There are two primary methods for character generation. One is
called the stroke method. This method creates characters out of a series of line seg-
ments, like strokes of a pen. We could build our own stroke-method character
generator by calls to the VECGEN algorithm. We would decide what line segments are
needed for each character and set up the calls to the VECGEN for each character we
wished to draw. In actual graphics displays, the commands for drawing the character
line segments may be in either hardware or software. The stroke method lends itself to
changes of scale; the characters may be made twice as large by simply doubling the
length of each segment.
The second method of character generation is the dot-matrix or bitmap method. In
this scheme, characters are represented by an array of dots. An array of 5 dots wide and
7 dots high is often used, but 7 X 9 and 9 x 13 arrays are also found. High-resolution
devices, such as ink-jet or laser printers, may use character arrays that are over 100
 GEOMETRY AND LINEGENERATION 23

FIGURE 1-12
Thick line construction.

pixels on a side. This array is like a small frame buffer, just big enough to hold a
character. The dots are the pixels for this small array. Placing the character on the
screen then becomes a matter of copying pixel values from the small character array
into some portion of the screen’s frame buffer (usually, for common alphanumeric ter-
minals the dot matrix is allowed to directly control the intensity of small parts of the
screen, eliminating the need for a large frame buffer). The memory containing the
character dot-matrix array is often a hardware device called a character-generator chip,
but random access memory may also be used when many fonts are desired. The size of
a dot is fixed, so the dot-matrix method does not lend itself to variable-sized charac-
ters. (See Figure 1-13.)

ee ieee Character generation. (a) Stroke method;

(b) (b) dot-matrix method.
24 CHAPTER ONE

Antialiasing techniques can be applied to characters. This can improve the ap-
pearance of the character, particularly for very small fonts and characters where the fi-
nite resolution of the display interferes with their smooth curved shapes.

DISPLAYING THE FRAME BUFFER

Using algorithms such as VECGEN or BRESENHAM, a frame buffer array may be |
modified to contain line segments and characters. This array directly corresponds to
the screen and holds an intensity-coded form of the image to be displayed. To use this |
device, some additional operations are required. We must be able to clear the frame
buffer. We need to begin with a blank array for the same reason that we start a pen-and-
ink drawing with a blank piece of paper. Another operation which is necessary is trans-
fer of the information from the frame buffer to the display medium (the display screen
or the paper). For raster graphics displays, this operation is built into the hardware. The
frame buffer is continually being scanned by the hardware to form the screen image.
Any change in the frame buffer is immediately shown on the screen. For graphics out-
put on a line printer, however, there is no hardware which automatically shows the con-
tent of the frame buffer. In this case, the display operation must be done by a software
routine. Finally, routines may be necessary for initialization and termination of the
graphics system. When starting a graphics program, hardware devices and storage may
be allocated and variables may be initialized by an initialization routine. Upon the
completion of the job, the deallocation of the storage, the release of hardware devices,
and other housekeeping chores may be done bya termination routine.
To complete this chapter, we would like to give the algorithms needed to obtain
graphics output from a line printer or common CRT terminal. The first algorithms, the
vector generators, have already been presented. We shall need a frame buffer array to
hold the image. The size of this array depends upon the resolution of the display de-
vice. If a CRT displaying 24 lines of 80 characters each is used, the frame buffer might
be dimensioned FRAME[80,24]. This is not the only choice. In many displays we find
that setting the lower right-hand pixel will cause automatic scrolling of the image,
shifting the top line off the screen. This can be prevented by not using the bottom line,
and [80,23] may be more appropriate. Or we may find that the display screen is not
square, and for the sake of treating the x and y directions equally, we may select a
square subarea like [60,23]. A line printer may use an array FRAME[90,50] or larger,
depending on the storage available and the number of characters per line. The frame
buffer will be an array of characters. The INTENSITY of algorithm 1.1 will be a
character such as the period or asterisk, out of which we will construct the picture. For
a clear or empty frame buffer corresponding to a clear display, the array should be
filled entirely with blanks. This is done by the following algorithm.

1.3 Algorithm ERASE Clears the frame buffer by assigning every pixel a background
value
Global FRAME the two-dimensional frame buffer array
WIDTH-START and HEIGHT-START the Starting indices of FRAME
WIDTH-END and HEIGHT-END the ending indices of FRAME
Local X, Y frame buffer indices
 GEOMETRY AND LINEGENERATION 25

BEGIN
FOR Y = HEIGHT-START TO HEIGHT-END DO
FOR X = WIDTH-STARTTO WIDTH-END DO
FRAME[X, Y] —*’;
RETURN;
END;

A call of the ERASE routine clears the display. We can use this whenever we wish to
draw a new picture.
The DISPLAY routine is used to show the contents of the frame buffer on aline
printer. Note that while our y coordinate begins at the bottom and increases as we
move up the display, printers typically begin at the top of the page and work down.
This means that the y coordinate must be displayed beginning with its high values and
working down.

1.4 Algorithm DISPLAY This displays the contents of the frame buffer
Global FRAME the frame buffer array
WIDTH-START and HEIGHT-START the starting indices of FRAME
WIDTH-END and HEIGHT-END the ending indices of FRAME
Local X,Y the pixel being displayed
BEGIN
FOR Y = HEIGHT-END TO HEIGHT-START DO
_ PRINT FOR X = WIDTH-START TO WIDTH-END, FRAME[X, Y];
RETURN;
END;

To complete this package, we must include a routine to initialize the parameters

for the display size and to perform any system-dependent housekeeping. Included in
the initialization should be the establishment of frame buffer size parameters and the
clearing of the display.

1.5 Algorithm INITIALIZE-1

Global WIDTH-START and HEIGHT-START the starting indices of FRAME
WIDTH-END and HEIGHT-END the ending indices of FRAME
WIDTH, HEIGHT the dimensions of FRAME
BEGIN
perform any needed storage allocation, hardware assignment,
or other system-dependent housekeeping;
HEIGHT-START < starting column index of FRAME;
WIDTH-START < starting row index of FRAME;
HEIGHT-END < ending column index of FRAME;
WIDTH-END < ending row index of FRAME;
HEIGHT <— HEIGHT-END — HEIGHT-START;
WIDTH < WIDTH-END — WIDTH-START,
ERASE;
RETURN;
END;
26 CHAPTER ONE

1.6 Algorithm TERMINATE

BEGIN
release any assigned hardware devices;
perform any necessary final*housekeeping;
STOP;
END;

FURTHER READING
There are a large number of texts on analytic geometry. One book which is oriented to-
ward computer graphics is [ROG76]. Use of a frame buffer was first described in
[NOL71]. Bresenham presented his algorithm in [BRE65]. DDAs and Bresenham’s al-
gorithm are also discussed in [EAR77], [FIE85], [LOC80], and [SPR82]. An alterna-
tive approach to vector generation based on the general form of the line equation rather
than on the parametric form is presented in [DAN70]. Another approach based on the
structural properties of the line is described in [BRO74] and [CED79]. A symmetric al-
gorithm which allows a sequence of lines to be erased by drawing them in reverse
order with ‘‘white ink’’ is given in [TRA82]. An example of a 7 X 9 dot-matrix font
may be found in [VAR71]. An example of specialized character-generation hardware
may be found in [THO72]. A discussion of antialiasing lines may be found in
[CRO78], [GUP81], and [PIT80]. A simple, fast algorithm for antialiased lines is
given in [KET85]. Antialiasing of characters is described in [WAR80]. A more general
discussion of aliasing and antialiasing is presented in [LEL80]. An early description of
CORE was published in [GSPC79]. An overview of the GKS standard is given in
[BON82]. The GKS standard is published in [GKS84].

[BON82] Bono, P. R., Encarnacao, J. L., Hopgood, FE R. A., ten Hagen, P. J. W., ‘“GKS—The First
Graphics Standard,’’ IEEE Computer Graphics and Applications, vol. 2, no. 7, pp. 9-23 (1982).
[BRE65] Bresenham, J. E., ‘‘Algorithm for Computer Control of a Digital Plotter,’’ JBM System Journal,
vol. 4, no.1, pp. 106-111 (1965).
[BRO74] Brons, R., “‘Linguistic Methods for the Description of a Straight Line on a Grid,’’ Computer
Graphics and Image Processing, vol. 3, no. 1, pp. 48-62 (1974).
[CED79] Cederberg, R. L. T., ‘‘A New Method for Vector Generation,’’ Computer Graphics and Image Pro-
cessing, vol. 9, no. 2, pp. 183-195 (1979). x
[CRO78] Crow, FE. C., ‘‘The Use of Grayscale for Improved Raster Display of,Vectors and Characters,”’
Computer Graphics, vol. 12, no. 3, pp. 1-5 (1978). ers
[DAN70] Danielsson, P. E., ‘‘Incremental Curve Generation,’’ JEEE Transactions on Computers, vol. C-19,
no. 9, pp. 783-793 (1970).
{[EAR77] Earnshaw, R. A., ‘‘Line Generation for Incremental and Raster Devices,’’ Computer Graphics,
vol. 11, no. 2, pp. 199-205 (1977).
[FIE85] Field, D., ‘Incremental Linear Interpolation,’? ACM Transactions on Graphics, vol. 4, no. 1, pp. 1-
11, (1985).
[GKS84] ‘‘Graphics Kernel System, ANSI X3H3/83-25r3,”’ Computer Graphics, vol. 18, special
issue (Feb.
1984).
[GSPC79] ‘‘Status Report of the Graphics Standards Planning Committee of ACM/SIGGRAPH,”’
Computer
Graphics, vol. 13, no. 3 (1979).
[GUP81] Gupta, S., and Sproull, R. FE, “Filtering Edges for Gray-Scale Displays,’’
Computer Graphics,
vol. 15, no. 3, pp. 1-5 (1981).
 GEOMETRY AND LINEGENERATION 27

[KET85] Ketcham, R. L., ‘‘A High-Speed Algorithm for Generating Anti-Aliased Lines,’’ Society for Infor-
mation Display 85 Digest, vol. 16, pp. 308-311 (1985).
[LEL80] Leler, W. J., ‘‘Human Vision, Anti-aliasing, and the Cheap 4000 Line Display,’’ Computer
Graphics, vol. 14, no. 3, pp. 308-313 (1980).
[LOC80] Loceff, M., ‘‘A New Approach to High-Speed Computer Graphics: The Line,’’ Computer, vol. 13,
no. 6, pp. 56-66 (1980).
[NOL71] Noll, A. M., ‘‘Scanned-Display Computer Graphics,’’ Communications of the ACM, vol. 14, no.
3, pp. 143-150 (1971).
[PIT80] Pitteway, M. L. V., and Watkinson, D. J., ‘‘Bresenham’s Algorithm with Grey Scale,’’ Communica-
tions of the ACM, vol. 23, no. 11, pp. 625-626 (1980).
[ROG76] Rogers, D. F, Adams, J. A., Mathematical Elements for Computer Graphics, McGraw-Hill, New
York (1976).
[SPR82] Sproull, R. F, ‘“Using Program Transformations to Derive Line-Drawing Algorithms,’’ ACM Trans-
actions on Graphics, vol. 1, no. 4, pp. 259-273 (1982).
[THO72] Thomas, P. A. V., and Mennie, W. E., ‘‘A Logic Character Generator for Use in a CRT Text Dis-
play,’’ Information Display, vol. 9, no. 2, pp. 9-i4 (1972).
[TRA82] Tran-Thong, ‘‘A Symmetric Linear Algorithm for Line Segment Generation,’’ Computers and
Graphics, vol. 6, no. |, pp. 15-17 (1982).
[VAR71] Vartabedian, A. G., ‘‘A Graphic Set for ASCII Using a7 x 9 Dot Pattern,’’ Information Display,
vol. 8, no. 6, pp. 11-16 (1971).
[WAR80] Warnock, J. E., ‘‘The Display of Characters Using Gray Level Sample Arrays,’’ Computer
Graphics, vol. 14, no. 3, pp. 302-307 (1980).

EXERCISES
1-1 Give the slope-intercept form of the equation for the line passing through each pair of
points.
a> (0, 0) and (6, 2)
o) (1, 0) and (7, 2)
c) (2, 3) and (4, 2)
d) (1, 1) and (2, 3)
e) (0, 2.5) and (-1, 3.5)
f) (3, 6) and (+4, 6)
1-2 Give a general form for the equation of the line passing through each pair of points in
Exercise 1-1.
1-3 Fof the following pairs of lines, state whether or not they intersect. If they do, give the coor-
dinates of the point of intersection.
a) y = xandy = 2x + 6
=x + 4andy = -2x-1
egme| x + 4andy = 2x + 6
d) y = -2x-landy = 2x + 6
e) y = -x/3 + 6andy = 2x + 6
f) y = -x/3 + 6andy = x/3 + 4
g) y = -x/3 + 6andy =x +4
h) y=xandy=x+4
i) y= xandy = x/3+4
j)y=-3andy =x+4
mid-
1-4 For each line segment specified by the following endpoints, give the midpoint and the
dle third line segment.
a) (0, 0) to (5, 0)
b) (1, 3) to (4, 7)
28 CHAPTER ONE

c)(4, 4) to (1, 7)
d) (1, 1) to (-3, 4)
e) (3, 5) to (-2, -7) : .
f)(2, 0) to (0, 2)
g) (2, 1) to (2, 4)
h) (2, 1) to (, 2)
1-5 What is the length of each line segment in Exercise 1-4?
c equations for x and y |
1-6 For each of the segments specified in Exercise 1-4, give parametri
with parameter u. The equations should give the first endpoint at u = 0 and the second endpoint |
atu = 1.
1-7 For the points

P= s(dyel)
P, = (4,5)
P; = (6, 2)

find the distance between each point and the line passing through the other two.
1-8 Consider the vectors
A = (3, 4]
B = [-1, 3]
C= le 12]

a) Give the length of each vector.

b) Find unit vectors for each of the three vector directions.
c) Show all possible pairwise additions of the vectors. |

1-9 Consider the vectors V, = [1, 4] and V, = [5, 2].

|
a) Plot the two vectors and their sum on graph paper such that each vector begins at the
origin.
b) Plot the locus of points (x, y) as u goes from 0 to 1 for
[x,y] =V=V, + uV>.

c) PlotV = V, + uV,
d) PlotV = uV, + (1-u)V>.
1-10 For a 10 X 10 frame buffer, interpret the BRESENHAM algorithm by hand to find which
pixels are turned on for the line segments between ~
..
a) (1, 2) and (7, 6)
b) (3, 2) and (6, 4)
c) (5, 8) and (9, 5)
1-11 The darkness of a line depends on the density of dark pixels for the line, that is, the
number of pixels set divided by the line’s true length. Calculate this density as a function of the
line’s slope for the BRESENHAM algorithm. Suggest ways to reduce this variation.
1-2 Because it only takes one bit to indicate if a pixel is on or off, we often find frame buffers
organized such that many pixels are packed into each computer word (one pixel for each bit in
the word). Suppose you had an array of 16-bit words for use as a frame buffer; write procedures
to set and examine individual pixels.
 GEOMETRY AND LINE GENERATION 29

PROGRAMMING PROBLEMS
1-1 Implement algorithms 1.1 or 1.2, and 1.3 through 1.6 to form a line-drawing system.
1-2 Test the system of Programming Problem 1-1 by writing a program to draw a house. For
example:
a) Call INITIALIZE-1 to set the frame parameters.
b) Call ERASE to clear the display.
c) Call the VECGEN or BRESENHAM algorithm to draw a house, such as
BRESENHAM (10, 5, 10, 10, ‘*’)
BRESENHAM (10, 10, 25, 15, ‘*’)
BRESENHAM (25, 15, 40, 10, ‘*’)
BRESENHAM (40, 10, 40, 5, ‘*’)
BRESENHAM (40, 5, 10, 5, ‘*’)
d) Call DISPLAY.
e) Call TERMINATE to end the program.

kkk kk
KK aKaKK
akk kkk
kkk kkk
kk kK
* *
* *
* *
* - *
KHRKKKEKKKRKKKKEKKKEREKEERERERERER

1-3 A widget manufacturing firm made the following sales:

Jan. 2000
Feb. 4000
Mar. 7000
Apr. 6000
May 4000
June 8000
“ July 8000
Aug. 7000
Sept. 5000
Oct. 6000
Nov. 8000
Dec. 9000

a graph of
Using the line-drawing system of Programming Problem 1-1, write a program to plot
the year’s widget sales. (See page 30, top.)
year’s
1-4 Using the widget data of Programming Problem 1-3, construct a bar graph of the
sales. (See page 31, top.)
The value of A
1-5 Write a routine to draw a square with a side length A centered on the display.
Test this routine by drawing a picture with four
should be given as an argument to the routine.
squares of different sizes. (See page 32, top.)
30 CHAPTER ONE

I
I
I
I : es
I ak

z
I
v
*

Tr eeKKKK *
iT: * ak *
I * * *
I * ak *
I * * * ak
I kkk * * *
I * * * * *
I * ae * * *
Tr * * kk ak *
I * * * * kk
tT ak * * x *
T * * * ee
I * kk Ok *
7 * zt &
aT * xk *
ip * wk
it * *
I *
T *
I *
I **
I*
I*
I*
*
£
E
I
I
I
I
I
[_——--—--—--——- --—-—-- ---

1-6 Write a program to draw an approximation to the curve x = y”/ 10. Do this by dividing the
y direction into intervals y,, y>,...,y,- Find the point on the curve corresponding to each x
value.

(x, y) = (y*/ 10, y) :

So for y; (y7/ 10, y,); for y2 (y3/ 10, y>); and so on. Connect these points with line segments to
get the approximation to the curve. (See page 32, bottom.)
1-7 Modify the BRESENHAM algorithm so that it will produce a dashed-line pattern. The dash
length should be independent of slope.
*1-8 Some CRT terminals allow positioning of the cursor by means of control characters or es-
cape sequences. If you have such a terminal, modify the vector generation algorithms so that
they alter the characters directly on the screen, instead of changing the value of a frame buffer.
*1-9 a) Modify the VECGEN algorithm to provide antialiasing. To simplify the problem, use
the vertical distance to the line (H and H — 1in the algorithm) to find intensities rather than the
 GEOMETRY AND LINEGENERATION 31

I
I
I
a
I *
e *
i *
I *
I * * * *
I * * * *
I * * * *
I * * * *
I * * * * * *
I * * * * * *
i * * * * * *
I * * * * * *
I * * * * * * * *
7 * * * * * * * *
I * * * * * * * *
T * * * * * * * *
T * * * * * * * * *
I * * * * * * * * *
I * * * * * * * * *
iT * * * * * * * * *
Tr * * * * * * * ® * * *
I * * * * * * * * * * *
I * * * * * * * * * * *
I * * * * * * * * * * *
Tt * * * * * * * * * * *
r * * * * * * * * * * *
I * * * * * * * * * * *
iT * * * * * * * * * * *
I * * * * * * * * * * *
I * * * * * * * * * * *
I WAT Fe * * * * * * * * *
I * * * * * * * * * * *
i * * * * * * * * * * *
I * * * * * * * * * * *
Tr * * * * * * * * * * *
I * * * * * * * * * * *
I * *————* ——— kk tk kkktt

true distance measured perpendicular to the line direction. Also assume that the intensity is prop-
ortional to 1 minus the distance. Test your algorithm on a scene containing lines at many differ-
ent angles.
b) The true distance between the line and the pixel center can easily be determined from
the vertical distance and the line’s slope. Note, however, that the sum of the perpendicular dis-
tances will not be 1, so they should not be used as a direct measure of pixel intensity. Instead, we
can imagine the line overlapping a portion of some extended pixel shape. The actual displayed
pixels will be brightest at their centers and then decline with distance, but they may overlap with
other pixels to provide uniform intensity areas. The spatial distribution of light from a pixel is
called its filter function. We can decide how much of this intensity distribution is claimed by the
line for some representative distances between line and pixel center, and store the result in a
table. This table provides the intensity for a pixel, given the distance between it and the line. Im-
plement this method of antialiasing.
c) We are spreading the line over two pixels instead of just turning on one, so the pixels
are gray, each representing part of the line and part of the background. But it can happen that
some of the background pixels are not ‘‘off’’ (e.g., when one line crosses another). To take this
into account, the intensity value determined for the new line should be combined with the exist-
32 CHAPTER ONE
ISIS IS IIIA
FERRI I III II II TIS IIIA IA ITA IAAI AISI III III IISA III SSSI
* *
* *
*
* HICH IIIA IIH IR III III AIK IIRIE II AIR IRR IERIEIAERERE RE
* * zi *
* * * *
* * * *
* * HHI IKI RRR EKER IAEA ERE ISIE EEREEEIN * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * kaRKKKRK KKK * * *
* * * * * * * *
* * * * * * * *
* * * * * * * *
* * * * * * * *
* * * * * * * *
* * * REKKKKKREKE * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * * * * *
* * HAAR IR IKI RIE REE ERE EER ERE REE EE * *
* * * *
* * * *
* * * *
* HERE HEIKERKE KEE REERERER EERE RE KEER ERE RRREEEERRERERREER *
* *
* *
HARKER ERK EERE ERERE HER RERE RE REREREREKEREEEREEREEERERERERERE

ing value, rather than replacing it. Develop a suitable combination scheme and modify your al-
gorithm to use it. Test it on a scene where lines cross.
*1-10 Construct a routine THICK-VECGEN(XA, YA, XB, YB, W, INTENSITY) which takes
two endpoints and a line width, and enters into the frame buffer a thick vector with width W cen-
tered on the line segment from (XA, YA) to (XB, YB).

ake
kkkKK
kkk
kee
kkk
eee
kk
ake
kkk
ae
kkk
ake
kk

ak
kk
ak
ae

*
 CHAPTER

GRAPHICS
PRIMITIVES

INTRODUCTION
The computer graphics user will not, in general, have to start from scratch in preparing
for a project. Instead, he will have a graphics system available. This system will in-
clude special hardware for output and input of pictorial information and software
routines for performing the basic graphics operations.
The purpose of a graphics system is to make programming easier for the user.
Just as high-level computer languages make programming easier by supplying power-
ful operations and constructs which match the requirements of the problem, a graphics
system supplies operations and constructs suited to the creation of graphical images to
enhance the development of a graphics program. In fact, some graphics systems are in
the form of special high-level graphics languages, languages suitable for solving
graphics problems. Other graphics systems are in the form of extensions to general-
purpose high-level languages such as FORTRAN, PL/I, or PASCAL. Such extensions
may be made through a “‘package’’ of subprograms or by the addition of new language
constructs.
While the form may vary between different graphics systems, there are certain
basic operations which can almost always be found. These are operations such as mov-
ing the pen (or electron beam), drawing a line, writing a character or a string of text,
and changing the line style. In this chapter we shall begin constructing our own
graphics system. We shall start by looking in more detail at some of the types of dis-
play devices. We shall present algorithms for some basic graphics operations, and we
shall introduce the concepts of device independence and ofa display file.

33
34 CHAPTERTWO

DISPLAY DEVICES
We saw in the last chapter how computer graphics images are composed of a finite
number of picture elements, or pixels. A display with good resolution might have 1000
divisions in both the x and y directions. The screen would then have 1000 x 1000 or 1
million, pixels. Each pixel requires at least one bit of intensity information, light or
dark, and further bits are needed if shades of gray or different colors are desired. Thus,
if we actually store the information for each pixel in the computer’s memory, a lot of
memory may be required. This is, in fact, what is done in some raster graphics dis-
plays. As we said in Chapter 1, the portion of the memory used to hold the pixels is
called the frame buffer. The memory is usually scanned and displayed by direct memory
access, that is, special hardware, independent of the central processor (leaving the pro-
cessor free for generation of the images). (See Figure 2-1.)
In the raster display, the frame buffer may be examined to determine what is cur-
rently being displayed. Surfaces, as well as lines, may be displayed on raster display
devices. Since images may be displayed on television-style picture tubes, raster dis-
play devices can often take advantage of the technological research and mass produc-
tion of the television industry. The raster terminal can also display color images. One
of the problems with the raster display is the time which may be required to alter every
pixel whenever the image is changed. Another disadvantage is the cost of the required
memory. This has been eased in some displays by using a coarse resolution (fewer
pixels). However, technological progress has steadily brought the price of memory
down, and at this point in time it appears that raster graphics will play a dominant role
in the future.
In the past, the cost of memory made the raster display seem much less promis-
ing. Different designs for graphics displays were developed in an effort to reduce ex-
penses. One approach was to let the display medium remember the image, instead of
using the computer memory. This is what plotters do. A pen is lowered onto paper and
moved under the direction of a vector generation algorithm. (See Figure 2-2.)
Once the line is drawn, the ink on the paper ‘‘remembers’’ it, and the computer
need not consider it further. The main disadvantage to this approach is that once
drawn, a line cannot be easily removed. If we wish to change the image on the plotter
by removing a line, we must get a fresh piece of paper and redraw the picture (without
the removed line). This can be time-consuming (and can use a lot of paper). For this
reason, plotters are not the best devices for interactive graphics>~
The first ‘‘low-cost’’ CRT display (under $5000) was produced by Tektronix. Be-
cause of good resolution and low cost, these terminals became widely accepted and
may be commonly found today. The terminals use special cathode ray tubes called di-
rect view storage tubes (DVST), which behave much the same way asa plotter. An elec-

LINE-ABS-2
(0.2, 0.2)

FIGURE 2-1
User program Frame buffer Display Raster display system.
 GRAPHICS PRIMITIVES 35

LINE-ABS-2
(0.2, 0.2)
FIGURE 2-2
User program Plotter Plotting system.

tron beam is directed at the surface of the screen. The position of the beam is con-
trolled by electric or magnetic fields within the tube. Once the screen phosphors of
this special tube have been illuminated by the electron beam, they stay lit. (See
Figure 2-3.)
As was the case with the plotter, one cannot alter a DVST image except by eras-
ing the entire screen and drawing it again. This can be done faster than on aplotter, but
the process is still time-consuming, making interaction difficult and eliminating these
devices from use in real-time animation.
A display device which stores the image (as plotters and storage tubes do) but al-
lows selective erasing is the plasma panel. The plasma panel contains a gas at low pres-
sure sandwiched between horizontal and vertical grids of fine wires. A large voltage
difference between a horizontal and vertical wire will cause the gas to glow as it does
in a neon street sign. A lower voltage will not start a glow but will maintain a glow
once started. Normally the wires have this low voltage between them. To set a pixel,
the voltage is increased momentarily on the wires that intersect the desired point. To
extinguish a pixel, the voltage on the corresponding wires is reduced until the glow
cannot be maintained. Plasma panels are very durable and are often used for military
applications. They have also been used in the PLATO educational system.
A device which is now becoming economical is the liquid crystal display. This is a
flat panel display technology, which makes it less bulky than CRTs. Also, because of
its low voltage and power requirements, it is lighter in weight, making it the display of
choice where portability is required. In a liquid crystal display, light is either transmit-
ted or blocked, depending upon the orientation of molecules in the liquid crystal. An
electrical signal can be used to change the molecular orientation, turning a pixel on or
off. The material is sandwiched between horizontal and vertical grids of electrodes
which are used to select the pixel.
At the present time, liquid crystal televisions are available, but the display is
formed on a single semiconductor wafer, which limits its size. Larger but slower dis-
plays are also available (640 x 250 pixels). This technology is still young and may
compete with CRT displays in the future.
Another approach to the display of graphical information, which has had a pro-
found effect on today’s graphics methods, is the vector refresh display. The vector re-

LINE-ABS-2
(0.2, 0.2)
FIGURE 2-3
User program DVST Display Direct view storage tube system.
36 CHAPTER TWO

fresh display stores the image in the computer’s memory, but it tries to be much more
efficient about it than a raster display. To specify a line segment, all that is required is
the coordinates of its endpoints. The raster frame buffer, however, stores not only the
endpoints but also all pixels in between, as well as all pixels not on the line. The vector
refresh display stores only the commands necessary for drawing the line segments. The
input to the vector generator is saved, instead of the output. These commands are
saved in what is called the display file. They are examined and the lines are drawn using
a vector-generating algorithm. This is done on a normal cathode ray tube, so the image
quickly fades. In order to present a steady image, the display must be drawn re-
peatedly. This means that a vector generator must be applied to all of the lines in an
image fast enough to draw the entire image before flicker is noticeable (more than 30
times a second). To do this, the vector generators of refresh displays are usually im-
plemented in hardware. (See Figure 2-4.)
Refresh displays do allow real-time alteration of the image. They also have some
disadvantages. The images formed are composed of line segments, not surfaces, and a
complex display may flicker because it will take a long time to analyze and draw it.
The concept of a display file has proved to be a useful one. It provides an inter-
face between the image specification process and the image display process. It also de-
fines a compact description of the image, which may be saved for later display. The
display-file idea may be applied to devices other than refresh displays. Such files are
sometimes called pseudo display files, or metafiles. Standards are currently under de-
velopment for metafiles to aid in the transport of images between computers.
Devices such as vector tefresh displays, pen plotters, and DVSTs may only di-
rectly support /ine drawing. They do not provide support for the solid areas which can
be constructed on raster displays. Such line-drawing devices are called calligraphic dis-
plays.
All the graphics display systems we have described can display the image as it is
being constructed. When a command to draw aline segment is issued, the line segment
can immediately appear on the screen or paper. The major difference is in whether the
display may be altered. To change a vector refresh display, one needs only to change
the display file. To change a raster display, one alters the frame buffer. A similar altera-
tion can change a plasma panel. However, to change a plotter or a DVST, one must
first call a new-frame routine, which shifts the paper or erases the screen, and then one
must redraw the entire image. a
We should mention one further class of display device. They are raster printers
and plotters that create the image in a single sweep across the page. There are a number
of technologies which fall into this category including film printers, laser printers,

LINE-ABS-2
(0.2, 0.2)

User program Display file Display processor Display

FIGURE 2-4
Vector refresh display system.
 GRAPHICS PRIMITIVES 37

electrostatic plotters, thermal and thermal-transfer printers, ink-jet printers, and impact
dot-matrix printers. The devices can range in resolution from about 100 pixels per inch
for dot-matrix printers to over 1000 pixels per inch for film printers. Dot-matrix print-
ers have an array of wires which can be individually triggered to press an inked ribbon
to make a dot on the paper. By sweeping the array across the paper, images can be
formed. Ink-jet printers form tiny droplets of ink which can be guided to the paper to
form dots. They use nozzles which move across the paper. Laser printers are built on
top of copier technology. Instead of copying the light pattern reflected from a piece of
paper, a laser is used to supply the light pattern. A rotating mirror sweeps the laser in a
raster pattern, and a light valve turns the beam on or off to form the image which is
““copied.*’ Film printers use a laser scanning system similar to laser printers, but they
focus the laser directly on photographic film to form the picture. Thermal printers have
a print head which can burn tiny dots on a heat-sensitive paper. Thermal-transfer print-
ers have a similar print head, but it is used to melt dots of wax-based pigment onto the
paper. Electrostatic plotters use an array of wires to which a voltage may be selectively
applied. As the paper passes across the wires, it is given a pattern of electrostatic
charges. The paper then passes through a liquid toner which is attracted to the pattern
on the paper.
As we saw in Chapter 1, one way to produce an image on such raster devices is
to first construct it in a frame buffer. The image is constructed by altering the contents
of this array, just as may be done for a raster display. The difference from araster dis-
play is that the user will not be able to watch the image being formed. When the user
has completed the image, it may output in the order required for printing. These de-
vices require an additional procedure for showing the image.
By creating a frame buffer, alphanumeric terminals and printers may be used for
graphics output. They have the same behavior as raster printing devices. The resolution
of such devices is usually not very good, but they are readily available and may be suf-
ficient for the user’s needs.

PRIMITIVE OPERATIONS
Regardless of the differences in display devices, most graphics systems offer a similar
set of graphics primitive commands (although the form of the commands may differ
between systems). The first primitive command we shall consider is that for drawing a
line segment. While a segment may be specified by its two endpoints, it is often the
case that the segments drawn will be connected end to end. (See Figure 2-5.)

FIGURE 2-5
Connected line segments.
 38 cCHAPTERTWO

The final point of the last segment becomes the first point of the next segment.
To avoid specifying this point twice, the system can keep track of the current pen or
electron beam position. The command then becomes: drawaline from the current posi-
tion to the point specified. This is

LINE-ABS-2(X, Y)

and is called an absolute line command because the actual coordinates of the final posi-
tion are passed. (See Figure 2-6.)
There is also a relative line command. In this command we only indicate how far
to move from the current position (see Figure 2-7):

LINE-REL-2(DX, DY)

The actual endpoint of the segment may be determined from the current position and
the relative specification. If we let (XC, YC) denote the current position, then

LINE-REL-2(DX, DY)

is the same as

LINE-ABS-2(XC + DX, YC + DY)

The above procedures are fine for producing a connected string of line segments, but it
may happen that we wish to draw two disconnected segments. (See Figure 2-8.)
This can be accomplished by the same mechanism if we picture these two seg-
ments as connected by a middle segment which happens to be invisible. We have com-
mands for moving the pen position without leaving a line. Again there can be both ab-
solute and relative moves.

MOVE-ABS-2(X, Y)
MOVE-REL-2(DX, DY)

We can construct a line drawing (say of a house) by a series of line and move
commands. If these commands are located in a subprogram, then each time a subpro-
gram is called, an image of the house is produced. If absolute commands
are used,
then the house will always be located at the same position on the screen.
If, however,

ye (X, Y)

FIGURE 2-6
The absolute line command.
 GRAPHICS PRIMITIVES 39

oy
HDX
{ FIGURE 2-7
The relative line command.

only relative commands are used, then the position of the house will depend upon the
position of the pen (or beam) at the time when the subprogram was entered. This may
be used for the construction of pictures made of repeated instances of basic compo-
nents. The subprogram for each type of component should be written using only rela-
tive commands. Drawing the entire picture is reduced to positioning the beam and call-
ing subprograms. For example, suppose we write the following subprogram for a
house.

Subprogram House
BEGIN
LINE-REL-2(0,0.2);
LINE-REL-2(0.1,0.1);
LINE-REL-2(0.1-0.1);
LINE-REL-2(0,-0.2);
_ LINE-REL-2(-0.2,0);
END;

This will start at the current pen position, which will become the lower-left corner of
the drawing. It will draw the left wall, the roof, the right wall, and, finally, the floor.
(See Figure 2-9.)
Since only relative commands were used, we can draw three houses by simply
calling this subprogram at three different starting positions. (See Figure 2-10.)

BEGIN
MOVE-ABS-2(0.1,0.2);
HOUSE;
MOVE-ABS-2(0.4,0.2);
HOUSE;
MOVE-ABS-2(0.7,0.2);
HOUSE;
END;

a FIGURE 2-8
Disconnected line segments.
40. cHaPTERTWO

FIGURE 2-9
Subprogram HOUSE.

THE DISPLAY-FILE INTERPRETER

While it might be possible to have the LINE command directly alter the frame buffer,
we shall not organize our system in this manner. Instead, we shall insert an inter-
mediate step. We shall have LINE and MOVE commands store their information in
what we shall call our display file. We shall then use the information in the display file
to create the image. There are several reasons for using this two-step process. First, not
every display device has a frame buffer. Different display devices require different pro-
grams to drive them. By isolating the driving program in the second of the two steps,
we achieve some measure of device independence. Second, it will make it easy for us
to change the position, size, and orientation of the image. These image transformations
will be carried out during the second step. The techniques involved will be covered in
Chapter 4. For now, we shall concentrate on the form of the display file. In Chapter 5
we shall learn how to structure the image and to carry out transformations on portions
of it. The display file will support this structuring.
The display file will contain the information necessary to construct the picture.
The information will be in the form of instructions such as ‘‘draw a line,’’ or ‘‘move
the pen.’’ Saving instructions such as these usually takes much less storage than saving
the picture itself. These instructions can be thought of as a program for creating the
image. Each instruction indicates a MOVE, or a LINE, action for the display device.
We shall write a display-file interpreter to convert these instructions into actual images.
(See Figure 2-11.)
Our interpreter may be thought of as a machine which executes these instruc-
tions. The result of execution is a visual image. In some graphics systems there is, in
fact, a separate computer, called the display processor, which is located in the graphics
terminal and which is used just for this purpose. In other systems, the behavior of a
display processor is simulated. Where there is a separate display )processor, some care
must be taken when the display file is modified since the display processor may cur-
rently be executing the instructions being changed.

FIGURE 2-10
Three calls of HOUSE, each with a different initial pen position.
 GRAPHICS PRIMITIVES 41

USER (PSEUDO)
PROGRAM INTERPRETER DISPLAY

FIGURE 2-11
Display file and interpreter.

Our display-file interpreter serves as an interface between our graphics program

and the display device. If we write a graphics program for a particular display device,
chances are that it will not run on a different display. The portability of the program is
limited. If, on the other hand, we write a program which generates display-file code,
all we need is an interpreter for each device which converts our “‘standard’’ display in-
structions to the actions of the particular device. We think of the display device and its
interpreter as a machine upon which any standard program may run. The display-file
instructions may actually be saved in a file either for display later or for transfer to
another machine. Such files of imaging instructions are sometimes called metafiles.

NORMALIZED DEVICE COORDINATES

Different display devices may have different screen sizes as measured in pixels. If we
wish our programs to be device-independent, we should specify the coordinates in
some units other than pixels and then use the interpreter to convert these coordinates to
the appropriate pixel values for the particular display we are using. The device-inde-
pendent units are called the normalized device coordinates. In these units, the screen
measures | unit wide and | unit high. The lower-left corner of the screen is the origin,
and the upper-right corner is the point (1, 1). The point (0.5, 0.5) is in the center of the
screen no matter what the physical dimensions or resolution of the actual display de-
vice may be. (See Figure 2-12.)
The interpreter uses a simple linear formula to convert from the normalized de-
vice coordinates to the actual device coordinates. Suppose that for the actual display
the index of the leftmost pixel is WIDTH-START and that there are WIDTH pixels in
the horizontal direction. Suppose also that the bottommost pixel is HEIGHT-START
and the number of pixels in the vertical direction is HEIGHT. In the normalized coordi-
nates the screen is 1 unit wide, but in the actual coordinates it is WIDTH units wide so
the normalized x position should be multiplied by WIDTH/I to convert to actual screen
units. At position x, = 0 in normalized coordinates we should get x, = WIDTH-
START in actual screen coordinates, so the conversion formula should be

(0.1) Ged)

FIGURE 2-12
(0, 0) (1,0) Normalized device coordinates.
42 cHAPTERTWO

x, = WIDTH « x, + WIDTH-START
Similarly for the vertical direction »

y, = HEIGHT« y, + HEIGHT-START
One problem in setting up the formula to convert from normalized to device
coordinates is that the display surfaces are often not square. The ratio of the height to
the width is called the display’s aspect ratio. If we have a display which is not square,
we can either use the display’s full height and width in the conversion formula or use
numbers which correspond to a square area. If we use the full dimensions, we take full
advantage of the display area, but the image will be stretched or squashed. If we use a
square area of the display, the image is correctly proportioned, but some of the display
area is wasted. If we use a square area larger than the actual display, we may use all of
the screen and have a properly proportioned image, but the image may not entirely fit
on the display.

DISPLAY-FILE STRUCTURE
Now let us consider the structure of the display file. Each display-file command con-
tains two parts, an operation code (opcode), which indicates what kind of command it is
(e.g., LINE or MOVE), and operands, which are the coordinates of a point (x, y). The
display file is made up of a series of these instructions. One possible method for stor-
ing these instructions is to use three separate arrays: one for the operation code (DF-
OP), one for the x coordinate (DF-X), and one for the y coordinate (DF-Y). To piece
together the seventh display-file instruction, we would get the seventh element from
each of the three arrays DF-OP[7], DF-X[7], and DF-Y[7]. The display file must be
large enough to hold all the commands needed to create our image.
We must assign meaning to the possible operation codes before we can proceed
to interpret them. At this point there are only two possible instructions to consider,
MOVE and LINE. We need to consider only absolute MOVE and LINE commands,
since relative commands can be converted to absolute commands before they are en-
tered into the display file. Let us define an opcode of 1 to mean a MOVE command
and an opcode of 2 to mean a LINE command. A command to move to position x =
0.3 and y = 0.7 would look like 1,.0.3, 0.7. The statements

DF-OP[3] <— 1;
DF-X[3] <— 0.3;
DF-Y [3] <— 0.7;

would store this instruction in the third display-file position. If DF-OP[4] had the value
2, DF-X[4] had the value 0.5, and DF-Y[4] had the value 0.8, then the display
would
show a line segment from (0.3, 0.7) to (0.5, 0.8) when display-file instruction
s 3 and
4 were interpreted. (See Figure 2-13.)
Let us develop the algorithms for inserting display-file instructions. Line seg-
ments require two endpoints for their specification, but we shall
enter only one end-
point and assume that the other endpoint is the current pen position.
We will therefore
 GRAPHICS PRIMITIVES 43

DF-OP DF-X DF-Y

Display file instructions Result

FIGURE 2-13
Display-file instructions.

need variables DF-PEN-X and DF-PEN-Y to keep track of the current pen position. We
will need to know this position for the conversion of relative commands to absolute
commands. We shall also need a variable FREE to indicate where the next free (un-
used) cell of the display file is located. These variables, together with the display file
itself, are used by several different routines and must be maintained between accesses.
They are therefore presented as global variables.
The first algorithm we will consider actually puts an instruction into the display
file.

2.1 Algorithm PUT-POINT(OP, X, Y) Place an instruction into the display file

Arguments OP, X, Y the instruction to be entered
Global DF-OP, DF-X, DF-Y the three display-file arrays
FREE the position of the next free cell
Constant DFSIZE the length of the display-file arrays
BEGIN
IF FREE > DFSIZE THEN RETURN ERROR ‘DISPLAY FILE FULL’;
DF-OP[ FREE] < OP;
DF-X[FREE] < X;
DF-Y [FREE] < Y;
BREE << FREE, 71;
- RETURN;
END;

This algorithm stores the operation code and the coordinates of the specified po-
sition in the display file. The pointer FREE to the next free cell is incremented so that
it will be in the correct position for the next entry.
We also wish to access elements in the display file. We isolate the accessing
mechanism in a separate routine so that any changes in the data structure used for the
display file will not affect the rest of our graphics package.

2.2 Algorithm GET-POINT (NTH, OP, X, Y) Retrieve the NTH instruction from the
display file
Arguments NTH the number of the desired instruction
OP. X, Y the instruction to be returned
Global arrays DF-OP, DF-X, DF-Y the display file
44 cuHaAPTERTWO

BEGIN
OP <— DF-OP[NTH];
X <— DF-X[NTH];
Y — DF-Y[NTH];
RETURN;
END;

Our MOVE and LINE instructions must update the current pen position and
enter a command into the display file. If the update of the pen position is done first,
then the new pen position will serve as the operand for the display-file instruction. It
will prove convenient in later chapters to have a separate routine which takes the opera-
tion code and the pen position and enters them into the display file as an instruction.

2.3 Algorithm DISPLAY-FILE-ENTER(OP) Combine operation and position to form

an instruction and save it in the display file
Argument OP the operation to be entered
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
PUT-POINT(OP, DF-PEN-X, DF-PEN-Y);
RETURN;
END;

Using DISPLAY-FILE-ENTER to place instructions in the display file, the abso-

lute MOVE routine becomes the following:

2.4 Algorithm MOVE-ABS-2(X, Y) User routine to save an instruction to move the pen
Arguments X, Y the point to which to move the pen
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
DF-PEN-X < X; |
DF-PEN-Y <— Y;
DISPLAY-FILE-ENTER(1);
RETURN;
END;

The point (DF-PEN-X, DF-PEN-Y) is keeping track of where we wish the pen to
go. By setting (DF-PEN-X, DF-PEN-Y) to (X, Y), we are saying the pen is to be at po-
sition (X, Y). ~

The algorithm for entering a LINE command is similar.

2.5 Algorithm LINE-ABS-2(X, Y) User routine to save a command to draw

a line
Arguments X, Y the point to which to draw the line
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
DF-PEN-X < X;
DF-PEN-Y < Y;
DISPLAY-FILE-ENTER(2)
RETURN;
END;
 GRAPHICS PRIMITIVES 45

Again by changing DF-PEN-X and DF-PEN-Y we indicate that the pen will be
placed at (X, Y), but by entering an operation code of 2 instead of 1, we instruct the in-
terpreter to drawaline as the pen is moved.
We can also write algorithms for the relative commands.

2.6 Algorithm MOVE-REL-2(DX, DY) User routine to save a command to move the
pen
Arguments DX, DY the change in the pen position
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
DF-PEN-X < DF-PEN-X + DX;
DF-PEN-Y <— DF-PEN-Y + DY;
DISPLAY-FILE-ENTER(1);
RETURN;
END;

2.7 Algorithm LINE-REL-2(DX, DY) User routine to save a command to drawa line
Arguments DX, DY the change over which to draw a line
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
DF-PEN-X <— DF-PEN-X + DX;
DF-PEN-Y <— DF-PEN-Y + DY;
DISPLAY-FILE-ENTER(2);
“ RETURN;
END;

The relative LINE and MOVE routines act like the absolute routines in that they
tell where the pen is to be placed and how it is to get there. They differ in that the new
pen position is calculated as an offset to the old pen position.

DISPLAY-FILE ALGORITHMS
The above algorithms indicate how to enter instructions into the display file. The other
half of the problem is to analyze the display file and perform the commands.
Now let us consider algorithms for our display-file interpreter. The interpreter
will read instructions from a portion of the display file and carry out the appropriate
LINE and MOVE commands by calls on a vector-generating subroutine such as those
described in Chapter 1.
The routine which actually causes the LINE or MOVE to be carried out on the
display can depend upon the nature of the display device and upon its software. Below
are versions of these algorithms which may be used with a vector generator. The
DOMOVE routine has only to update the current pen position. The arithmetic involved
is the conversion from normalized device coordinates to the actual screen coordinates.

2.8 Algorithm DOMOVE(X, Y) Perform a move of the pen

Arguments X, Y point to which to move the pen (normalized coordinates)
Global FRAME-PEN-X, FRAME-PEN-Y the pen position (actual screen coordi-
nates)
46 CHAPTERTWO

WIDTH, HEIGHT the screen dimensions

WIDTH-START, HEIGHT-START coordinates of the lower-left corner
WIDTH-END, HEIGHT-END coordinates of the upper-right corner
BEGIN
FRAME-PEN-X < MAX(WIDTH-START, MIN(WIDTH-END, X * WIDTH +
WIDTH-START));
FRAME-PEN-Y <— MAX(HEIGHT-START, MIN(HEIGHT-END, Y « HEIGHT +
HEIGHT-START));
RETURN;
END;

In this algorithm we see the formula for converting the normalized coordinate
values of the arguments into actual screen coordinates. The MAX and MIN functions
have been added to the formula as a safeguard. They prevent it from ever generating a
value outside the bounds of the actual display. If (X, Y) were to correspond to a point
outside the screen area, the MAX and MIN functions would ‘‘clamp’’ the correspond-
ing screen coordinate position to the display boundary. In algorithm 2.8, it was as-
sumed that WIDTH-START was less than WIDTH-END, and HEIGHT-START less
than HEIGHT-END; however, this is not the case for some devices. In general, we use
the smallest boundary in the MAX test and the largest boundary in the MIN test.
The DOLINE algorithm updates the pen position and calls the Bresenham al-
gorithm (or some other vector generator) to place the line segment in the frame buffer.
It, too, performs a conversion from normalized device coordinates to the actual screen
coordinates. Since the Bresenham algorithm is used in this example, the coordinates
must be rounded to integer pixel positions.

2.9 Algorithm DOLINE(X, Y) This routine draws a line

Arguments X, Y point to which to draw the line (normalized coordinates)
Global FRAME-PEN-X, FRAME-PEN-Y the pen position (screen coordinates)
WIDTH, HEIGHT the screen dimensions
WIDTH-START, HEIGHT-START coordinates of the lower-left corner
WIDTH-END, HEIGHT-END coordinates of the upper-right corner
LINECHR the style of the line
Local X1, Y1 the old endpoint of the line segment
BEGIN
X1 < FRAME-PEN-X; ee
Y1 <— FRAME-PEN-Y; ‘.
FRAME-PEN-X <- MAX(WIDTH-START, MIN(WIDTH-END, X * WIDTH +
WIDTH-START));
FRAME-PEN-Y <— MAX(HEIGHT-START, MIN(HEIGHT-END, Y * HEIGHT +
HEIGHT-START));
BRESENHAM(INT(X1 + 0.5), INT(Y1 + 0.5),
INT(FRAME-PEN-X + 0.5), INT(FRAME-PEN-Y + 0.5),
LINECHR);
RETURN;
END;

Now we can write the interpreter, the routine which examines

the display file and
calls DOMOVE and DOLINE according to the instructions
that it discovers. It will
 GRAPHICS PRIMITIVES 47

prove useful in later chapters to be able to interpret just a portion of the display file.
We
shall therefore pass as arguments to the interpreter the starting position START and
how many instructions to interpret COUNT.

2.10 Algorithm INTERPRET(START, COUNT) Scan the display file performing the
instructions
Arguments START the starting index of the display-file scan
COUNT the number of instructions to be interpreted
Local NTH the display-file index
OP, X, Y, the display-file instruction
BEGIN
a loop to do all desired instructions
FOR NTH = STARTTO START + COUNT- 1 DO
BEGIN
GET-POINT(NTH, OP, X, Y);
IF OP = 1 THEN DOMOVE(X, Y)
ELSE IF OP = 2 THEN DOLINE(X, Y)
ELSE RETURN ERROR ‘OP-CODE ERROR’;
END;
RETURN;
END;

A loop steps through all the desired display-file instructions, retrieving them by
the GET-POINT routine. The algorithm identifies each instruction as being either a
MOVE or a LINE instruction by examining its opcode and calls the routine DOMOVE
or DOLINE to actually perform the appropriate action upon the display.

DISPLAY CONTROL
In order to show the picture described in the display file, one might have to do three
things. First, the current display may have to be cleared; second, the display file must
be interpreted; and third, on some devices (such as line printers and standard CRT ter-
minals) an explicit action is required to show the contents of the frame buffer. For the
convenience of the user we shali combine these operations under a single routine. Be-
fore presenting the algorithm for this routine, however, we should sayalittle more
about clearing the display (or frame buffer). We may not want to clear the display
every time we interpret the display file. If the only changes to the picture have been ad-
ditions, then there is no need to clear and redraw the rest of the image. In fact, quite
complex drawings may be generated a piece at a time. Of course, some changes made
to the image do require clearing and redrawing. Deleting a portion of the drawing —or
shifting its position, size, or orientation—may require a new image starting with a
clear screen. We shall see how to perform these changes in Chapters 4 and 5; the im-
portant thing to note here is that the request for a clear screen may be incorporated as
part of these changes. In these cases, screen clearing can be done automatically. Still,
the user should also be able to explicitly request an erasure of the screen. In Chapter 1
we introduced a display-clearing routine which the user might call; however, some-
times the point at which the user’s program discovers that an erasure is needed may
48 cHAPTERTWO

occur before the desired time at which the actual clearing should occur. We shall han-
dle clearing of the frame by means of a flag (a variable with the value true or false).
We use a true value to indicate that the screen should be cleared before interpreting the
display file and a false value to mean that the display-file instructions may be drawn
‘‘on top of’’ the old image. If the machine discovers that an erasure will be needed, it
sets the flag to true. If the user decides that there should be a new frame, he sets the
flag to true. Nothing happens to the display until the display file is ready to be inter-
preted. At this point the erase flag is checked, and if true, the frame is cleared and the
flag is changed back to false.

2.11 Algorithm NEW-FRAME User routine to indicate that the frame buffer should be
cleared before showing the display file
Global ERASE-FLAG aflag to indicate whether the frame should be cleared
BEGIN
ERASE-FLAG < TRUE;
RETURN;
END;

Now we wish to combine the erasing of the frame buffer, if needed, the interpre-
tation of the display file, and the displaying of the new frame buffer into a single
routine called MAKE-PICTURE-CURRENT.

2.12 Algorithm MAKE-PICTURE-CURRENT User routine to show the current dis-

play file
Global FREE the index of the next free display-file cell
ERASE-FLAG indicates ifframes should be cleared
BEGIN
IF ERASE-FLAG THEN
BEGIN
ERASE;
ERASE-FLAG < FALSE;
END;
IF FREE > 1 THEN INTERPRET(1, FREE - 1);
DISPLAY;
FREE:<1;
RETURN;
END;

The algorithm first checks for an erasure, as we have discussed. It next checks to
be sure the display file is not empty; if it is not, the commands within it are interpreted.
Next, any actions necessary to show the results of the interpretation are taken by
means
of the DISPLAY routine. If a graphics display device which immediately shows
the re-
sult of each command is used, then the call of DISPLAY is unnecessar
y or DISPLAY
becomes a “‘dummy”’ do-nothing routine. Finally, the display-file index
is reset to 1,
indicating an empty display file ready to accept the next image-drawing
commands.
There is one more routine needed to implement this stage of
our graphics sys-
tem. We have some global variables which should be given
initial values. This is done
by the following algorithm.
 GRAPHICS PRIMITIVES 49

2.13 Algorithm INITIALIZE-2A Initialization of variables for line drawings

Global FREE the index of the next free display-file cell
DF-PEN-X, DF-PEN-Y the display-file pen position
BEGIN
INITIALIZE-1;
PREE <— Je
DF-PEN-X < 0;
DF-PEN-Y < 0;
NEW-FRAME;
RETURN;
END;

TEXT
Another primitive operation is that of text output. Most graphics displays involve tex-
tual, as well as graphical, data. Labels, instructions, commands, values, messages,
and so on, may be presented with the image. The primitive command involved here is
the output of a character or a string of characters. While one usually has a command
available for output of an entire string, there may sometimes be only a command avail-
able for output of a single character. A procedure to apply such a primitive to an entire
string is not difficult to construct. The characters themselves may be drawn by either
the dot-matrix or the stroke method. Their patterns are often copied from memory into
the frame buffer or created by special character generation hardware, although
software’ subroutines using line segments are also to be found. The advantage of
hardware is speed and a saving of display-file memory. With sophisticated displays
there may be options which the user must specify. Among such options are the spacing
of the characters, the size of the characters, the ratio of their height to their width, the
slope of the character string, the orientation of the characters, and, possibly, which
font is to be used.
We will extend our interpreter to include the output of text. We will do this by ex-
tending the number of operation codes to include one code for each character. The
operand for an instruction will determine where the character will be placed on the
screen. We shall use as the opcode for a character the negative of its ASCII character
code. (See Figure 2-14.) Our opcodes will range between —32 and —126 inclusive.
Using the ASCII character code should facilitate conversion between the instruction
and the character value for output in most systems. We shall exclude the ASCII codes
for control characters so that a valid code must be less than —31 (the codes 0 through
—31 will be used for line and polygon style changes). Character codings other than
ASCII can be used, provided that the codes can be mapped to values less than —31.
A character command, then, has an opcode that specifies which character is to
be displayed and x, y operands which tell where on the screen that character should be
placed. A word or string of text is stored as a sequence of individual character instruc-
tions. Since we save the position of each character, we have control over the character
spacing within the string. Since we save y position as well as x position, we can cause
our strings to be written vertically or diagonally.
We have not introduced any mechanism for specifying the character orientation.
Some display devices allow characters to be drawn at any angle. Others allow only 90-
50 cHAPTERTWO

32 8 56 P 80 h 104
re 33 9 57 Q 81 i 105
y 34 : 58 R 82 j 106
# 35 . 59 S 83 k 107
$ 36 < 60 7 84 } 108
% 37 = 61 U 85 m 109
& 38 > 62 V 86 n 110
39 2 63 W 87 ° 111
( 40 @ 64 x 88 P 112
) 41 A 65 Y 89 q 113
* 42 B 66 ie 90 r 114
¥. 43 Cc 67 [ 91 s 115
f 44 D 68 \ 92 t 116
= 45 E 69 } 93 u 117
46 F 70 ‘A 94 V 118
/ 47 G 71 zs 95 w 119
0 48 H 72 96 x 120
1 49 I 73 a 97 y 121
2 50 J 714 b 98 z 122
3 51 K 75 c 99 { 123
4 52 L 16 d 100 124
5 53 M 77 e 101 } 125
6 54 N 78 f 102 ~ 126
7 55 O 79 g 103
FIGURE 2-14
ASCII character codes.

degree increments, while many (such as normal output devices) display only upright
characters. To maintain device independence and to try to simplify things a little, we
shall require all characters within a string to be oriented in the same direction and, in
fact, will design our algorithms for a device which allows only upright characters.
Writing a string of text can be rather cumbersome if each character requires its own
procedure call. We shall therefore write a routine for placing entire strings of text into
the display file through a series of character commands. It will be up to this routine to
automatically shift the position of each character to achieve the desired spacing. The
spacing between characters will be given by global variables XCHRSP and YCHRSP.
Of course, we need a routine to set these spacing parameters to whatever values we de-
sire. The CORE and GKS graphics systems have rich selections of text formatting op-
erations; they allow changing the size, orientation, spacing, and font of the character
and also the direction of the line of text. We will not be that ambitious here. We shall
only control the character spacing and line direction. We shall determine the direction
of the line of text from the orientation of the characters. Theté is a command SET-
CHARUP(DX, DY) in which [DX, DY] is a vector specifying the ‘‘UP”’ direction for
the characters. The text line then prints to the “‘right’’ of this direction. For our sys-
tem, we will not try to change character orientation with respect to line direction, so a
SET-CHARUP command looks a bit awkward. Nevertheless, we shall use it because
of its correspondence to graphics standards.
Spacing in the direction of the text line can be changed by the SET-
CHARSPACE(CHARSPACE) command. The system will space over one character
width automatically. The CHARSPACE parameter indicates any additional spacing.
This spacing is specified in units of character size, so a CHARSPACE of 0.5 would
mean spacing an additional one-half character (a total of 1.5 character widths between
character centers). The CHARSPACE parameter is measured in terms of character
 GRAPHICS PRIMITIVES 5]

width instead of normalized device coordinates, so a widely spaced line on one device
will look just as widely spaced on every other device, even though the width of the
characters (in normalized device coordinates) may differ. Our problem is to convert the
CHARUP and CHARSPACE specifications, and our knowledge about the size of a
character, into distances to step in the x and y direction between each character in the
string (XCHRSP, YCHRSP). (See Figure 2-15 and Figure 2-16.)
The calculation of step size is complicated by the fact that we are not rotating the
character symbols and that the width and height of a character are often unequal. If the
text string is to be printed horizontally, then we shall step by the character width. If
the string is to be printed vertically, then character height should be the default step
size. At other orientations, we wish to use some default step size between the width
and height. To use as much of the width as we are stepping horizontally and as much of
the height as we are stepping vertically, we calculate a default step size as
|(CHAR-WIDTH)(DY) | + |(CHAR-HEIGHT)(DX) |
DEFAULT-STEP =
(DX? 5 + DY?)ane
(2.1)

Here DX and DY are from the CHARUP vector, and are perpendicular to the line direc-
tion; that is why they appear to be reversed.
This default step size can be increased by the user’s CHAR-SEPARATION fac-
tor.

TRUE-STEP = DEFAULT-STEP (1 + CHAR-SEPARATION) (2.2)

Finally, we decompose the step size into horizontal and vertical components.
Once again, since [DX, DY] is a vector perpendicular to the direction of the string, DY
will act as the x component and — DX will behave as the y component.
DY
XCHRSP = TRUE-STEP
(2.3)
—DxX
YCHRSP = TRUE-STEP

This calculation is carried out a bit more efficiently in the following algorithm.
(You may notice that upon combining Equations 2.1, 2.2, and 2.3, the square root op-
eration drops out.)

CHAR-WIDTH

CHAR-HEIGHT
\ CHARUP

CHAR-SEPARATION

FIGURE 2-15
Text parameters for orientable char-
acters.
52 cHAPTERTWO

CHAR-WIDTH

CHAR-HEIGHT

CHAR-SEPARATION

FIGURE 2-16
Text parameters for nonorientable
characters.

2.14 Algorithm SET-CHARUP(DX, DY) User routine to indicate in which direction a

string should be printed
Arguments DX, DY a vector for the up direction of the string
Global XCHRSP, YCHRSP spacing between character centers
CHAR-WIDTH, CHAR-HEIGHT the character size
CHAR-SEPARATION the proportion of character size separating characters
Local S, S1, S2 temporary variables used to hold partially completed calculations
Constant ROUNDOFF some small number greater than any round-off error
BEGIN
S<—DX 2:4 DY ff2;
IF S < ROUNDOFF THEN RETURN ERROR ‘NO CHARUP DIRECTION’;
S1 <— ((CHAR-WIDTH * DY| + |CHAR-HEIGHT « DX));
S2 —S1 «(1 + CHAR-SEPARATION) / S;
XCHRSP < DY « S2;
YCHRSP < — DX x S82;
RETURN;
END;

The SET-CHARSPACE routine should not only change the global CHAR-
SEPARATION value but also update the step size currently in use. In the arguments of
the call to SET-CHARUP, we again see the switch of x and y which comes from
specifying the “‘up’’ direction in terms of the parameters for the’text line direction.
~

2.15 Algorithm SET-CHARSPACE(SPACE) User routine to set the spacing between

characters
Argument SPACE the fraction of the character size separating characters
Global CHAR-SEPARATION storage for the character spacing
XCHRSP, YCHRSP spacing between character centers
BEGIN
CHAR-SEPARATION < SPACE;
SET-CHARUP( — YCHRSP, XCHRSP);
RETURN;
END;
 GRAPHICS PRIMITIVES 53

Now let’s write the algorithm to enter a string of text into the display file, begin-
ning at the current pen position.

2.16 Algorithm TEXT(STRING) User routine to place instructions for printing a string
into the display file
Argument STRING the characters to be entered
Global DF-PEN-X, DF-PEN-Y the pen position
XCHRSP, YCHRSP the character spacing
Local LEN the length of the string.
X, Y the character position
I an index to count off the characters
CHR the character being saved
OP the character’s operation-code
BEGIN
determine the length of the string
LEN <— LENGTH(STRING)
save the pen position for restoration after string is entered
X < DF-PEN-X;
Y <— DF-PEN-Y;
enter the string
FOR I = 1 TO LEN DO
BEGIN
consider the ith character of the string
CHR <— GETCHR(STRING, J);
form its character code
OP <— —- DECODE(CHR);
enter it into the display file
DISPLAY-FILE-ENTER(OP);
move the pen to the next character position
DF-PEN-X < DF-PEN-X + XCHRSP;
DF-PEN-Y <- DF-PEN-Y + YCHRSP;
END;
restore the pen to its original position
MOVE-ABS-2(X, Y);
-RETURN;
END;

The routines LENGTH, GETCHR, and DECODE depend upon how strings are
represented on the particular system that we are using, and the algorithms for them will
not be presented here. The LENGTH routine returns the number of characters in the
string. The GETCHR routine returns the ith character in the string. The DECODE
routine converts the character to the ASCII code, if necessary.
We must now modify our interpreter to be able to handle these new character
commands which we are entering into the display file.

2.17 Algorithm INTERPRET(START, COUNT) (Algorithm 2.10 revisited) Scan the

display file performing the instructions
Arguments START the starting index of the display-file scan
COUNT the number of instructions to be interpreted
54 CHAPTERTWO

Local NTH the display-file index

OP, X, Y the display-file instruction
BEGIN s
a loop to do all desired instructions
FOR NTH = START TO START + COUNT- 1 DO
BEGIN
GET-POINT(NTH, OP, X, Y);
IF OP < -31 THEN DOCHAR(OP, X, Y)
ELSE IF OP = 1 THEN DOMOVE(X, Y)
ELSE IF OP = 2 THEN DOLINE(X, Y)
ELSE RETURN ERROR ‘OP-CODE ERROR’;
END;
RETURN;
END;

The instructions for actually putting the character into the frame buffer have
been placed in the DOCHAR routine.

2.18 Algorithm DOCHAR(OP, X, Y) Place a character on the screen

Arguments OP indicates which character should be used
X, Y indicate the position on the screen
Global WIDTH, HEIGHT the screen dimensions
WIDTH-START, HEIGHT-START screen coordinates of lower-left corner
WIDTH-END, HEIGHT-END screen coordinates of upper-right corner
Local CHR the character to be displayed
X1, Y1 screen coordinates of the character
BEGIN
CHR < CODE(-OP);
X1 < MAX(WIDTH-START, MIN(WIDTH-END, X * WIDTH +
WIDTH-START));
Y1 — MAX(HEIGHT-START, MIN(HEIGHT-END, Y * HEIGHT +
HEIGHT-START));
GENERATE-CHAR(X1, Y1, CHR);
RETURN;
END;

Once again we have system-dependent routines, which are ‘named CODE and
GENERATE-CHAR. CODE converts from ASCII to whatever form is convenient for
GENERATE-CHAR. GENERATE-CHAR generates a character on the screen or in the
frame buffer at position (X1, Y1).
Again we have introduced some global variables which should be initialized.
The following routine takes care of this.

2.19 Algorithm INITIALIZE-2B Initialization of character parameters

Global CHAR-WIDTH, CHAR-HEIGHT the character size
CHAR-SEPARATION the proportion of character size to use for additional
character spacing
 GRAPHICS PRIMITIVES 55

BEGIN
INITIALIZE-2A;
CHAR-WIDTH <- the width of a character in normalized coordinates;
CHAR-HEIGHT < the height of a character in normalized coordinates;
CHAR-SEPARATION < 0;
SET-CHARUP(O, 1);
RETURN;
END;

THE LINE-STYLE PRIMITIVE

Many display devices offer a selection of line styles. Lines may be continuous, or they
may be dashed or dotted. One may be able to select the color of the line or its intensity
or thickness. It is desirable to be able to change the line style in the middle of the dis-
play process. We therefore need a display-file command for changing line style. When
the interpreter encounters such a command, the line style is changed and all sub-
sequent lines are drawn in this new style. Our display-file commands are composed of
three parts, the opcode and the two operands for the x and y coordinates. We can use a
special opcode to indicate change of line style (or color or intensity), but such a com-
mand would not require any operands. Some possible display-file organizations allow
different-sized instructions, but we shall take the simpler alternative of providing
dummy operands which are ignored. We will use codes between 0 and —15, inclusive,
for change of line-style commands. This allows up to 16 possible line styles. The ac-
tual identification of opcodes with line styles will depend upon the possible line styles
available. We shall, however, assume that a code of 0 corresponds to the normal
straight line. This should be the default style when the system is initialized. Other line
styles should correspond to the codes —1, —2, and so on. For aline printer display such
as discussed in Chapter 1, the line style is the character that is placed in the frame buf-
fer. Changing the line style is a matter of changing this character. (See Figure 2-17.)
We will augment our interpreter to process commands for changing line style. To
do this, we will need two new routines, one to place the line-style commands into the
display file and another to actually make the change whenever such a command is dis-
covered by the interpreter.
While line-style opcodes in the display file are negative to distinguish them from
the MOVE and LINE opcodes, there is no reason to force the user to use negative num-
bers in his specification. To do so would give the user one more pointless thing to re-
member. We therefore have the user specify the positive codes 1,2,... for line styles

FIGURE 2-17
Changing line style.
56 CHAPTERTWO

and let the SET-LINESTYLE algorithm enter corresponding negatives into the display
file. The algorithm for setting the line style would look as follows:

2.20 Algorithm SET-LINESTYLE(LSTYLE) User routine for changing line style

Argument LSTYLE the user’s line-style specification
BEGIN
DISPLAY-FILE-ENTER(1 — LSTYLE);
RETURN;
END;

When the interpreter discovers a line-style command, it must perform the

change. We isolate this system-dependent process in a separate routine to allow easy in-
terfacing with different display devices.

2.21 Algorithm DOSTYLE(OP) Routine to change the line style

Argument OP indicates the desired line style
BEGIN
decode OP;
set the line style;
RETURN;
END;

If the vector generation algorithms of Chapter 1 are used for a line printer or
CRT with different characters representing different line styles, then algorithm 2.21
would be

2.21A Algorithm DOSTYLE(OP) Routine to change the line style

Global LINECHR the line character used by the vector generator
BEGIN
LINECHR < LINE-CODE(1 — OP);
RETURN;
END;

where LINE-CODE converts from the integers 1, 2, ... to appropriate character codes ’

forexampley 7 ort na.

We must once again extend our interpreter so that it will now recognize line-style
operation codes.

2.22 Algorithm INTERPRET(START, COUNT) (Algorithm 2.17 revisited) Scan the

display file performing the instructions
Arguments START the starting index of the display-file scan
COUNT the number of instructions to be interpreted
Local NTH the display-file index
OP, X, Y the display-file instruction
BEGIN
a loop to do all desired instructions
FOR NTH = START TO START + COUNT- 1 DO
 GRAPHICS PRIMITIVES 57

BEGIN
GET-POINT(NTH, OP, X, Y);
IF OP < -31 THEN DOCHAR(OP, X, Y)
ELSE IF OP < 1 THEN DOSTYLE(OP)
ELSE IF OP = 1 THEN DOMOVE(X, Y)
ELSE IF OP = 2 THEN DOLINE(X, Y)
ELSE RETURN ERROR ‘OP-CODE ERROR’;
END;
RETURN;
END;

We shall also make a small change in the DISPLAY-FILE-ENTER algorithm so

that it will distinguish between instructions which use the coordinate information
(LINE and MOVE) and those which do not (STYLE). This distinction will prove use-
ful in later chapters, where we will not enter all of the line-drawing instructions but
will still wish to enter all changes of style.

2.23 Algorithm DISPLAY-FILE-ENTER(OP) (Algorithm 2.3 revisited) Combine op-

eration and position to form an instruction and save it in the display file
Argument OP the operation to be entered
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
IF OP < 1 AND OP > -32 THEN PUT-POINT(OP, 0, 0)
ELSE PUT-POINT(OP, DF-PEN-X, DF-PEN-Y);
RETURN;
END;

There is one more routine which is needed to complete this stage of our system.
We must supply initial or default values to our global variables. In addition to the pre-
vious initializations, we should now also set the initial line style.

2.24 Algorithm INITIALIZE-2 Initialization of variables for lines, characters, and

style
“ BEGIN
INITIALIZE-2B
DOSTYLE(0)
RETURN;
END;

AN APPLICATION
Let us suppose that we wanted a graphics program to draw a graph of some data. Let
us outline how this might be done using some of the algorithms which we have writ-
ten. To begin, we can plot the horizontal and vertical axes of the graph. These are just
two lines, which may be created by a MOVE-ABS-2 to one endpoint and a LINE-
ABS-2 to the other endpoint. (See Figure 2-18.)
58 cHaPTERTWO

/
f FIGURE 2-18
Drawing two axes.

MOVE-ABS-2(0.2, 0.8);
LINE-ABS-2(0.2, 0.2);
LINE-ABS-2(0.8, 0.2);

The next step might be to label the axes. This can be done bya series of TEXT
commands. For example, suppose we wish to label the horizontal axis with the num-
bers | through 5. For each number, we decide where it should be placed on the graph.
We issue a MOVE-ABS-2 to place the pen at this position, followed by a TEXT com-
mand to write the numeral. (See Figure 2-19.)

MOVE-ABS-2(0.3, 0.15);
TEXT(‘1’);
MOVE-ABS-2(0.4, 0.15);
TEXTC2))s

We might also want to label each axis with a string indicating what is being plot-
ted. A MOVE-ABS-2 to the starting position followed by a TEXT command will do
this for us. (See Figure 2-20.)

ave ee FIGURE 2-19

2 Bee 4 5 Labeling an axis.
 GRAPHICS PRIMITIVES 59

a
1 2 3 4 5 FIGURE 2-20
TIME IN SECONDS More labeling.

MOVE-ABS-2(0.35, 0.1);
TEXT(‘TIME IN SECONDS’);

For the vertical axis, it may be useful to change the CHAR-UP direction so that
the print line will be vertical. (See Figure 2-21.)

SET-CHARUP(-1, 0);
MOVE-ABS-2(0.1, 0.3);
TEXT(‘POSITION IN METERS’);

Now all that is left for us to do is to plot the data. At this point we might use
SET-LINESTYLE to change the style of the line so that our data curve looks different
from the axes.
We shall assume that the information to be displayed came into our plotting pack-
age as an array of data values or measurements. For each of these measurements we
may have to perform some arithmetic to calculate the corresponding position on our
display. If we use MOVE-ABS-2 to plot the first data point, and LINE-ABS-2 on the
remaining points, we shall get a sequence of line segments connecting the data values.
(See Figure 2-22.)

SET-LINE-STYLE(2);
MOVE-ABS-2(0.2, A[1] + 0.2);

wu
ES
wwn

POSITION
METERS
IN —

‘STE ee FIGURE 2-21

TIME IN SECONDS Labeling with a different CHAR-UP.
60 cHaPTERTWO

METERS
IN
POSITION

1 2 3 4 5 FIGURE 2-22
TIME IN SECONDS Plotting the curve with a different line style.

FOR I = 2TO6DO
LINE-ABS-2(0.1 «I + 0.1, A[I] + 0.2);

Plotting of all but the first data point can usually be accomplished by a loop
which gets the measurement, calculates the corresponding display position, and calls
LINE-ABS-2 to display it.

FURTHER READING
Introductory discussions of various imaging technologies may be found in [ALD84}j,
[BAL84], [HOB84], [HUB84], [JER77], [LEE84], [LUC78], [MCC84], [MIL84],
([PRE78], [SLO76], [SPR83], [WAT84], [ZAP75], and [ZUC84]. Use of magnetic
drum memory as a frame buffer was described in [OPH68], and the first example of
the core memory frame buffer is [NOL71]. A frame buffer is also described in
[KAJ75]. A display processor and display-file interpreter is described in [DOO84].
Use of a display file for a device-independent system and suggested primitives are pre-
sented in [NEW74]. A discussion of graphics primitives is given in [LUC76], and
those in the CORE system in [BER78] and [GRA79].

[ALD84] Aldersey-Williams, H., ‘‘Liquid Crystals in Flat Panel Displays,’’ Electronic Imaging, vol. 3, no.
11, pp. 54-57 (1984). .
[BAL84] Baltazzi, E. S., ‘‘Reprographic Imaging Techniques,’’ Electronic Imaging, vol. 3, no. 12, pp. 53-
55 (1984). :
[BER78] Bergeron, R. D., Bono, P. R., and Foley, J. D., ‘‘Graphics Programming Using the CORE Sys-
tem,’? ACM Computing Surveys, vol. 10, no. 4, pp. 389-394 (1978).
[DOO84] Doomink, D. J., Dalrymple, J. C., ‘‘The Architectural Evolution of a High-Performance
Graphics Terminal,’’ JEEE Computer Graphics and Applications, vol. 4, no. 4, pp. 47-54 (1984).
[GRA79] Graphic Standards Planning Committee, ‘‘Status Report Part II: General Methodology and the Pro-
posed Core System,’’ Computer Graphics, vol. 13, no. 3, pp. II-1-II-179 (1979).
[HOB84] Hobbs, I. C., ‘“Computer Graphics Display Hardware,’’ Computer Graphics and Applications, vol.
1, no. 1, pp. 25-32 (1981).
[HUB84] Hubbold, R. J., ‘‘Computer Graphics and Displays,’’ Computer-Aided Design, vol. 16, no. 3, pp.
127-133 (1984).
[JER77] Jern, M., ‘‘Color Jet Plotter,’’ Computer Graphics, vol. 11, no. 1, pp. 18-31 (1977).
 GRAPHICS PRIMITIVES 61

[KAJ75] Kajiya, J. T., Sutherland, I. E., Cheadle, E. C., ‘‘ A Random-Access Video Frame Buffer,’’ Pro-
ceedings of the Conference on Computer Graphics, Pattern Recognition, & Data Structure, pp. 1-6,
IEEE Cat. No. 75CH0981-1c (1975).
[LEE84] Lee, F C., Mills, R. N., Talke, F E., ‘‘The Application of Drop-on-Demand Ink Jet Technology to
Color Printing,’’ JBM Journal of Research and Development, vol. 28, no. 3, pp. 307-313 (1984).
[LUC76] Lucido, A. P., ‘Software Systems for Computer Graphics,’’ Computer, vol. 9, no. 8, pp 23-36
(1976).
{LUC78] Lucido, A. P., ‘‘An Overview of Directed Beam Graphics Display Hardware,’’ Computer, vol. 11,
no. 11, pp. 29-37 (1978).
[MCC84] McComrrick, J. J., ‘‘Present and Future Color Display Technologies for Graphics,’’ Computers &
Graphics, vol. 8, no. 3, pp. 281-293 (1984).
[MIL84] Miller, R. C., ‘‘Introduction to the IBM 3800 Printing Subsystem Models 3 and 8,’’ JBM Journal
of Research and Development, vol. 28, no. 3, pp. 252-256 (1984).
[NEW74] Newman, W. M., and Sproull, R. FE, ‘‘An Approach to Graphics System Design,’’ Proceedings of
the IEEE, vol. 62, no. 4, pp. 471-483 (1974).
{[NOL71] Noll, A. M., ‘‘Scanned-Display Computer Graphics,’’ Communications of the ACM, vol. 14, no.
3, pp. 143-150 (1971).
{OPH68] Ophir, D., Rankowitz, S., Shepherd, B. J., and Spinrad, R. J., ‘‘BRAD: the Brookhaven Raster
Display,’’ Communications of the ACM, vol. 11, no. 6, pp. 415-416 (1968).
([PRE78] Preiss, R. B., ‘‘Storage CRT Display Terminals, Evolution and Trends,’’ Computer, vol. 11, no. 11,
pp. 20-28 (1978).
[SLO76] Slottow, H. G., ‘‘Plasma Displays,’’ JEEE Transactions on Electron Devices, vol. ED-23, no. 7, pp.
760-772 (1976).
([SPR83] Spruth, W. G., Bahr, G., ‘‘Printing Technologies,’’ Computers and Graphics, vol. 7, no. 1, pp. 51-
57 (1983).
[WAT84] Watkins, H. S., Moore, J. S., ‘‘A Survey of Color Graphics Printing,’’ JEEE Spectrum, vol. 21,
no. 7, pp. 26-37 (1984).
[ZAP75] Zaphiropoulos, R., ‘‘Electrostatic Impactless Printing and Plotting,’’ Jnformation Display, vol. 12,
no. 1, pp. 23-25 (1975).
[ZUC84] Zuckerman, M. M., ‘‘Innovative Display Technologies— Why a Flat Panel When You Can Have a
CRT?,’’ Computer Graphics & Applications, vol. 4, no. 4, pp. 9-15 (1984).

EXERCISES
2-1 Suppose that the display device we are using has resolution 1024 x 1024. The pixels are
numbered from 0 to 1023 for both the x and y directions, so0 = x, = 1023.
- a) Give the conversion function f(x,) for converting from normalized coordinates x, to
the corresponding screen coordinates x,, that is
X, = f(x,)

b) If the pixels were numbered from 1 to 1024 (1 = x, = 1024), what would the
conversion function be?
2-2 Write program fragments using LINE-ABS-2 and MOVE-ABS-2 to draw each of the
following pictures.

(2) (0) (c) (d) (e)

 62 CHAPTERTWO

FIEIA eae
2-3 Assuming that the pen is initially positioned at (0, 0) for each case, write program frag-
ments to draw the pictures of Exercise 2-2 using the LINE-REL-2 and MOVE-REL-2 routines.
2-4 Give the call to SET-CHARUP for generating each of the following line orientations.

rereEXA
orY
—+
(b) a
><
—

=ip) ~
OR

—><M—
(@) X4ee Nedag et ace|
alll

TEXT

he
2-5 Write a single subprogram to draw the symbol

~<——————
0.3 ———_—_—__>

using only relative commands. The entire symbol should be 0.1 unit high and 0.3
unit wide. By
calling the subprogram three times, construct the picture:

4 HA HG
 GRAPHICS PRIMITIVES 63

2-6 Using the routines described in this chapter, write program fragments to generate the
following displays.

tt
a) +5 VOLTS

2-7 Suppose that MOVE and LINE display-file instructions each take 4 bytes of storage.
a) How much display-file storage is required for 1000 independent line segments?
b) If the line segments are drawn as connected sequences that average four segments in
length, how much display-file storage is needed for 250 such sequences?
c) If the image is placed in a 1024 x 1024, 1-bit-per-pixel frame buffer, how much frame
buffer memory is required?
2-8 Give the advantages and disadvantages of hard copy displays, DVST displays, vector re-
fresh displays, and raster displays. Suggest an applications area for which each class of device is
best suited.
2-9 A high-speed laser printer may print two 8.5 X 11 inch pages each second at a resolution of
600 pixels per inch. How many bits per second does such a device require?
2-10 If a TV screen has 525 scan lines and an aspect ratio of 3:4, and if each pixel contains 8
bits worth of intensity information, how many bits per second are required to show 30 frames
each second?
2-11 We have presented a display-file structure in which each entry is the same size. However, a
more flexible and space-efficient scheme might allow different instructions to be different sizes
(e.g., style instructions smaller than line instructions) or it might even allow an instruction to
vary in size (e.g., string instructions instead of character instructions). Design a display-file
structure which accommodates dynamic instruction size.
2-12 Discuss the advantages of a display-file structure which contains instructions for entire
strings over a structure which contains only individual character instructions, when one wishes
to interpret the display file on several different devices.

PROGRAMMING PROBLEMS
2-1 a) Implement algorithms 2.1 through 2.13.
b) Implement algorithms 2.14 through 2.19.
c) Implement algorithms 2.20 through 2.24.
64 cCHAPTERTWO

2-2 Test the routines of Programming Problem 2-1 by drawing a house.

a) Use all of LINE-ABS-2, LINE-REL-2, MOVE-ABS-2, and MOVE-REL-2 to
construct an image of a house. .
b) Test the character routines by tabeBhg the FLOOR with a horizontal string, the WALL
with a vertical string, the ROOF with a diagonal string, and the entire HOUSE with a widely
spaced string.
c) Draw the roof of the house with a different line style.

FLOOR
rors
HOUSE

2-3 Write the four routines

INQUIRE-LINESTYLE(L) INQUIRE-CHARSIZE(W, H)
INQUIRE-CHARSPACE(S) INQUIRE-CHARUP(DX, DY)
which return to the user the current settings of the line style, character separation, character
width and height, and the last specified character up direction.
2-4 Write routines called
POLYLINE-ABS-2 (X-ARRAY, Y-ARRAY, N)
POLYLINE-REL-2 (DX-ARRAY, DY-ARRAY, N)
These routines will enter a connected sequence of lines. The sequence begins at the current pen
position, and for POLYLINE-ABS-2 connects the sequence of absolute points stored in X-
ARRAY and Y-ARRAY. For POLYLINE-REL-2, the sequence is given by relative changes
stored in DX-ARRAY and DY-ARRAY. The number of values in the arrays is given by N.
2-5 a) Using the LINE and MOVE routines of this chapter, or the POLYLINE routines of
Programming Problem 2-4, draw aline graph of the following widget sales data.
Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec
2000 4000 7000 6000 4000 8000 8000 7000 5000 6000 8000 9000

THOUSANDS
IN
SALES

JF M AM JU So OuNT D
MONTH
 GRAPHICS PRIMITIVES 65

b) Draw and label the axes for the widget graph.

c) Use solid lines for the axes and some other line style for the graph itself.
2-6 Construct a bar graph of the widget data given in Programming Problem 2-5 using LINE
and MOVE routines.

IN
SALES
THOUSANDS

VEAP SSM) SAT MI 1 Acs S OF" N4D

MONTH

2-7 a) Write three user subprograms called

SET-MARKER-SYMBOL(SYMBOL)
MARKER-ABS-2(X, Y)
MARKER-REL-2(DX, DY)
The SET-MARKER-SYMBOL routine saves a character or symbol to be used by the other two
routines. The two MARKER subprograms move the pen without drawing and then place the
marker symbol at this position. Thus the commands
SET-MARKER-SYMBOL(2);
MARKER-ABS-2(0.2, 0.3);
MARKER-REL-2(0.2, 0.0);
would result in the picture

b) Use the marker routines to plot the widget sales information of Programming Problem
2-5. (See page 66, top.)
c) Write routines called
POLYMARKER-ABS-?2 (KX-ARRAY, Y-ARRAY, N)
POLYMARKER-REL-2 (DX-ARRAY, DY-ARRAY, N)
These routines would draw the current marker symbol at each of the points indicated by the
argument arrays. For POLYMARKER-ABS-2, the arrays X-ARRAY and Y-ARRAY contain a
66 CHAPTER TWO

IN
SALES
THOUSANDS

J Fo MSA MeIR aI Ae IS-8O> (ND

MONTH

sequence of absolute points. For POLYMARKER-REL-2, the points are described by a

sequence of relative position changes in DX-ARRAY and DY-ARRAY. The number of values in
the arrays is given by N.
d) Use the POLYMARKER-ABS-2 and the POLYMARKER-REL-2 routines to plot the
widget sales information of Programming Problem 2-5.
2-8 Write an automatic graphing procedure which takes an array of values and plots them on a
graph
a) Using markers or character symbols to plot points
b) Using connected line segments
2-9 Write routines called
DASHED-ABS-2(X, Y)
DASHED-REL-2(DX, DY)
These procedures enter a sequence of line segments into the display file so that the result is the
display of a dashed line.
2-10 a) Write a routine called
ADJUSTED-TEXT(STRING, LENGTH)
which takes a string as an argument and also a length. The routine will find the number of
characters in the string, examine the character width (perhaps using INQUIRE-CHARSIZE of
Programming Problem 2-3), and call SET-CHARSPACE before — such that the
displayed string has the requested length.
b) Using the ADJUSTED-TEXT routine, write a program which reads a stream of text,
breaks it up into lines, if necessary, and displays those lines in a column which has both left and
right margins justified.
*2-11 The projection of three-dimensional constructions and the display of surfaces will be
considered in Chapters 8 and 9; however, in this problem we shall consider an easy way to
present three-dimensional information using only the two-dimensional routines which we have
studied. Consider a function of two variables w = f(u,v). We can set up three coordinate axes
for u, v, w, and plot the (u, v, w) points in space.
The collection of these points forms a surface. We can get some feel for the shape of this
surface by plotting sample points and connecting them by line segments.
i
GRAPHICS PRIMITIVES 67

To place the three-dimensional point on our two-dimensional display, note that as w in-
creases the point moves up, as u increases the point moves right, and as v increases the point
moves both right and up.

x =u
+ v/2
y=we v/2

Now as an example of this technique, consider the following array of w values

0.1 0.10 0.10 0.13 0.15 0.20

0.2 0.10 0.12 0.15 0.20 0.25
0.3 0.12 0.15 0.20 0.25 0.30
Ora Ot 0.15 0.20 0.25 0.30
0.5 0.08 0.11 0.15 0.18 0.20

a) Write a program to plot the table points using the above formula to determine the
appropriate (x, y) coordinates.

. ®

s - e e
e
rn e
a e
r e e
e
e e ®
e
@
e e @ ©
e ®

b) Connect the successive elements of each row by line segments.

68 CHAPTERTWO

c) Connect with line segments successive row elements and successive column elements.

*2-12 To get three-dimensional realism in a display of a function of two variables, we should

not draw lines that would be hidden by the surface. The following is a simple but rather crude
method to get this effect. We wish to show a function such as w = f(u, v). As in Programming
Problem 2-11, we map this into our x and y coordinates by

x =u
+ v2

y=we v/2

For given v value, we shall draw curves showing how w varies as u is changed. We do this by
stepping through u values; calculating w by the function f; using u, v, and w to get x and y; and
connecting successive x and y points by line segments. To get the hidden-line effect, we must
draw the curves in front first. That means we start with the smallest v, draw the curve for it, in-
crement v, draw the next curve, and so on. Now suppose we draw a curve which rises to form a
peak. This peak might hide the next curve. If this is the case, the y value for the curve in front
will be greater than the y value for the curve behind. We can avoid drawing the hidden line if in-
stead of drawing the line segment to the new y value, we draw it to the maximum yvalue. To do
this, we just need an array which holds maximum values. The following is an algorithm which
uses this method to draw the function for the range 0 < u < 0.5, 0 < v < 0.5; NUM-CURVES
is the number of curves drawn, MAX-TABLE is the array of maximum y values and should con-
tain at least three times NUM-CURVES elements.

BEGIN
initialize the array of maximum values
FOR I = 1TO 3 * NUM-CURVES DO MAX-TABLE[I] < 0;
set the step sizes
DV —0.5 / (NUM-CURVES - 1);
DU < DV « 0.5;
V<—0:
loop to draw all curves
FOR T = 1 TO NUM-CURVES DO
BEGIN
UE 10:
loop to draw one curve
FOR K = 1 to 2 * NUM-CURVES - 1 DO
BEGIN
calculate a point on the curve
W < F(U, V);
 GRAPHICS PRIMITIVES 69

X—U+ V0.5;
Ne Wey e095;
plot the point
MAX-TABLE [T + K - 1] — MAX(MAX-TABLE[T + K - 1], Y)
IF K = 1 THEN MOVE-ABS-2(X , MAX-TABLE[T + K - 1)
ELSE LINE-ABS-2(X, MAXTABLE [T + K -1]);
step along the curve
Oe DU:
END;
step to next curve
V=V+ DV:
END;
RETURN;
END;

Try it for f(u, v) = (1/2) (sin 87u) (sin 4rv) forO = u $0.5, 0 Sv < 0.5. The illustration was
done with NUM-CURVES at a value of 25.

*2-13 Revise the graphics system's display-file structure so that it can handle instructions of
varying size. Using this display file, revise the text algorithms so that they use a string instruc-
tion instead of individual characters.
*2-14 It is possible to have a single graphics program driving several different display devices
or workstations. The program can build a single display file which is examined by different inter-
preters for the different devices. Extend your graphics system to allow multiple workstations
(perhaps some CRTs and some hard-copy devices). Provide the user the ability to select which
workstations will be active at any time.
 CHAPTER

THREE: )
POLYGONS

INTRODUCTION
So far our discussion has dealt only with lines and characters. The world might seem
rather dull if it were made out of straight lines. It would be a world of stick figures and
outlines lacking texture, mass, and solidity. How much more interesting is the world of
patterns, colors, and shading. Unfortunately, much of the early graphics dealt only
with line drawings. This was because the available display devices (plotters, DVSTs,
and vector refresh displays) were line-drawing devices. Raster displays, however, can
display solid patterns and objects with no greater effort than that involved in showing
their outlines. Coloring and shading are possible with raster technology. Thus with the
rise in popularity of raster displays has also come an increase in attention to surface
representation. The representation of surface objects is important even for line-drawing
displays. In Chapter 9 we shall look at the problem of not showing lines which would
normally be hidden, such as the edges on the back side of an object. For this problem,
knowledge of the edges is not enough; we must also know what surfaces are present so
that we can decide what can and what cannot be seen. :
In this chapter we will extend our system to include a new graphic primitive, the
polygon. We shall discuss what polygons are and how to represent them. We shall learn
how to determine if a point is inside a polygon. Finally, a method for filling in all in-
side pixels will be developed.

POLYGONS
We wish to be able to represent a surface. Our basic surface primitive is a polygon (a
many-sided figure). A polygon may be represented as a number of line segments con-

70
 potycons 71

nected end to end to form aclosed figure. Alternatively, it may be represented as the
points where the sides of the polygon are connected. The line segments which make up
the polygon boundary are called sides or edges. The endpoints of the sides are called the
polygon vertices. The simplest polygon is the triangle, having three sides and three ver-
tex points. (See Figure 3-1.)
We can divide polygons into two classes: convex and concave. A convex polygon
is a polygon such that for any two points inside the polygon, all points on the line seg-
ment connecting them are also inside the polygon. A concave polygon is one which is
not convex. A triangle is always convex. (See Figure 3-2.)
Kan

POLYGON REPRESENTATION
If we are to add polygons to our graphics system, we must first decide how to represent
them. Some graphics devices supply a polygon drawing primitive to, directly image
polygon shapes. On such devices it is natural to save the polygon as a unit. Other de-
vices provide a trapezoid primitive. Trapezoids are formed from two s¢an lines)and two
line segments. (See Figure 3-3.) The trapezoids are drawn by steppingdown the line
segments with two vector generators and, for each scan line, filling in all the pixels be-
tween them. Every polygon can be broken up into trapezoids by means similar to those
described below. In such a system, it would be natural to represent a polygon as a
series of trapezoids. Many other graphics devices do not provide any polygon support
at all, and it is left up to the software to break up the polygon into the lines or points
which can be imaged. In this case, it is again best to represent the polygon as a unit.
This is the approach we shall take. We shall store full polygons in our aay Lie and.
investigate the methods for imaging them using lines and points. = “7 ~~
What should a polygon look like in our display file? We might just ice line
commands for each of the edges into the display~file, but there are two deficiencies
with this scheme. First, we have no way of knowing which line commands out of all
those in the display file should be grouped together to form the polygon graphical unit;
and second, we have not correctly positioned the pen for drawing the initial side. We
can overcome these problems by prefacing the commands for drawing the polygon
sides by a new command. This new command will tell us how many sides are in the
polygon, so we will know how many of the following line commands are part of the
polygon. Upon interpretation, this new command eal act like a move to correctly posi-
tion the pen for drawing the first side.
We have not yet used the operation ¢odes 3 or greater. We can therefore use these
codes to indicate polygons. The value of the code will indicate the number of sides. We

Porn «A
FIGURE 3-1
Polygons.
CoN ald Se
Convex polygons Concave polygons

FIGURE 3-2
Convex and concave polygons.

will therefore be limited to polygons with no more sides than the maximum possible
opcode. The X and Y operands of the polygon command will be the coordinates of the
point where the first side to be drawn begins. Since polygons are closed curves, it will
also be the final endpoint of the last side to be drawn. Upon execution, the polygon in-
struction (opcode 3 or greater) will signal that the following instructions belong to a
polygon, but will otherwise behave as a move command. (See Figure 3-4.)

ENTERING POLYGONS
We can now consider algorithms for entering polygons into the display file. The infor-
mation required to specify the polygon is the number of sides and the coordinates of
the vertex points. Arrays can be used to pass the coordinates of all vertices to the

am
—
FIGURE 3-3
A polygon can be drawn asaseries of trapezoids.
 POLYGONS 73

(0.3, 0.6)

(0.1,,0.4) (0.5, 0.4)

DF-OP DF-X DF-Y

(0.1, 0.1) (0.3, 0.1)

FIGURE 3-4
A polygon and its display-file entry.

routine. If we give absolute coordinates, the following algorithm may be employed.

(See Figure 3-5.)

3.1 Algorithm POLYGON-ABS-2(AX, AY, N) Entry of an absolute polygon into the

display file
Arguments AX, AY arrays containing the vertices of the polygon
; N the number of polygon sides
Global DF-PEN-X, DF-PEN-Y the current pen position
Local I for stepping through the polygon sides
BEGIN
IF N < 3 THEN RETURN ERROR ‘POLYGON SIZE ERROR’;
enter the polygon instruction
DF-PEN-X <— AX[N];
DF-PEN-Y <— AY[N];
DISPLAY-FILE-ENTER(N);
enter the instructions for the sides
FOR I = 1TON DO LINE-ABS-2(AX{I], AY[I]);
RETURN;
END;

We might also wish to be able to construct polygons relative to the current pen
position. We would then understand the first point specified to be a relative move from

(AX[1], AY[1])
a

a
Z
Ne
a
On
a

Pe

FIGURE 3-5
(AX(3], AY([3]) (AX[2], AY[{2]) |Absolute polygon.
74 CHAPTER THREE

the current position, and subsequent points to be used in relative line commands for
the sides. (See Figure 3-6.)

3.2. Algorithm POLYGON-REL-2(AX, AY, N) Entry of a relative polygon into the

display file
Arguments AX, AY arrays containing the relative offsets of the vertices
N the number of polygon sides
Global DF-PEN-X, DF-PEN-Y the current pen position
Local I for stepping through the polygon sides.
TMPX,TMPY Storage of the point at which the polygon is closed.
BEGIN
IF N < 3 THEN RETURN ERROR ‘POLYGON SIZE ERROR’;
DF-PEN-X < DF-PEN-X + AX[1];
DF-PEN-Y <— DF-PEN-Y + AY[1];
save the starting point for closing the polygon
TMPX < DF-PEN-X;
TMPY < DF-PEN-Y;
enter the polygon instruction
DISPLAY-FILE-ENTER(N);
enter the instructions for the sides
FOR I = 2TON DO LINE-REL-2(AX{I], AY[I]);
close the polygon
LINE-ABS-2(TMPX, TMPY);
RETURN;
END;

In the above algorithm, the polygon starting position must be calculated from the
specified offset and the current pen position. This location must be saved temporarily
so that it may be used in the final instruction, which closes the figure.

AN INSIDE TEST
Having entered the commands in the display file, we can show outlined forms for poly-
gons by simply modifying the interpreter so that it treats command codes 3 and greater
as move commands. However, we might also wish to be able to show the polygons as
solid objects by setting the pixels inside the polygon as well as those on the boundary.
Let us consider how we can determine whether or not a point is inside of a polygon.
One method of doing this is to construct a line segment between the point in question

FIGURE 3-6
{[AX[3], AY[3]] Relative polygon.
 potycons 75

and a point known to be outside the polygon. It is easy to find a point outside the poly-
gon; one could, for example, pick a point with an x coordinate smaller than the small-
est x coordinate of the polygon’s vertices. One then counts how many intersections of
the line segment with the polygon boundary occur. If there are an odd number of inter-
sections, then the point in question is inside; an even number indicates that it is out-
side. This is called the even-odd method of determining polygon interior points. (See
Figure 3-7.)
When counting intersection points, one must be cautious when the point of inter-
section is also the vertex where two sides meet. To handle this case, we must look at
the other endpoints of the two segments which meet at this vertex. If these points lie
on the same side of the constructed line, then the point in question counts as an even
number of intersections. If they lie on opposite sides of the constructed line, then the
point is counted as a single intersection. (See Figure 3-8.)
There is an alternative method for defining a polygon’s interior points called the
winding-number method. Conceptually one can stretch a piece of elastic between the
point in question and a point on the polygon boundary. The end attached to the polygon
is slid along the boundary until it has made one complete circuit. We then examine the
point in question to see how many times the elastic has wound around it. If it has
wound at least once, then the point is inside. If there is no net winding, then the point
is outside. Calculating the winding number for a point is not as difficult as the method
just described. We begin, as in the even-odd method, by picturing a line segment run-
ning from outside the polygon to the point in question and consider the polygon sides
which it crosses. However, in the winding-number method, instead of just counting the
intersections, we give each boundary line crossed a direction number, and we sum these
direction numbers. The direction number indicates the direction the polygon edge was
drawn relative to the line segment we constructed for the test. For example, to test a
point (x,, y,), let us consider a horizontal line segment y = y, which runs from out-
side the polygon to (x,, y,). We find all of the sides which cross this line segment.
Now there are two ways for a side to cross. The side could be drawn starting below the
line, cross it, and end above the line (first y value less than the second y value). In this
case, we give a direction number of —1 to the side. Or the edge could start above the
r

FIGURE 3-7
Even-odd inside test.
76 CHAPTER THREE

Counts odd

Counts even

FIGURE 3-8
Count of vertex intersections.

line and finish below it (first y value greater than the second y value). This case is
given a direction of 1. The sum of the direction numbers for the sides that cross the
constructed horizontal line segment yields the winding number for the point in ques-
tion. (See Figure 3-9.)
Using the winding number to define the interior points can yield different results
from the even-odd method when a polygon is allowed to overlap itself. (See Figure
3-10.) The polygon-filling algorithm presented in this book is based on the even-odd
method.

POLYGON INTERFACING ALGORITHMS

Before we become too deeply involved in the details of filling a polygon, let us give
the algorithm that is needed to interface polygons with the rest of our graphics system.
We shall be able to show polygons either filled or in outline. We shall therefore
provide the user with a method of indicating his preference. This is done by setting a
global flag which can be checked at the time when the polygon is actually drawn.

FIGURE 3-9
Calculating winding number as the sum of
the direction numbers for the sides crossed.
 POLYGONS 77

Even-odd method Winding number method

FIGURE 3-10
Interior points for a polygon that overlaps itself.

3.3 Algorithm SET-FILL(ON-OFF) User routine to set a flag indicating that polygons
should be filled
Argument ON-OFF the user’s fill setting
Global SOLID a flag which indicates filling of polygons
BEGIN
SOLID < ON-OFF;
RETURN;
END;

Just as we have different line styles, we shall provide for different interior styles
for polygons. Fill styles may be implemented as different colors, different shades of
gray, or different filling patterns. We shall give the user a routine for selecting the de-
sired interior style. Style values between | and 16, inclusive, will be acceptable. We
map these values to opcodes between —16 and —31.

y- 3.4 Algorithm SET-FILL-STYLE(STYLE) User routine to set the polygon interior

style
Argument STYLE the user’s style request
Constant FIRST-FILL-OP first fill op = —16
BEGIN
DISPLAY-FILE-ENTER( FIRST-FILL-OP — (STYLE — 1));
RETURN;
END;

Fill styles are handled in the same manner as line styles. A code indicating the
desired style setting is placed in the display file. We can use negative integers as the
operation codes for both line and fill styles. In our system, the codes 0 through —15
will be reserved for line styles and codes —16 through —31 will refer to fill styles. The
user again specifies a positive integer for the interior style, which is converted to a
negative number between —16 and -31 before it is stored in the display file. What the
78 CHAPTER THREE

system actually does with a polygon fill style depends upon how fill styles are im-
plemented. The system might select'a color or an intensity value (FILLCHR) to be
used in filling. Or it might set a SCAN-DECREMENT parameter which indicates the
filling of every nth scan line. It might also set an index into a table of patterns (FILL-
PATTERN).
The DOSTYLE algorithm of Chapter 2 can be modified to include both line and
interior style settings.

3.5 Algorithm DOSTYLE(OP) (Algorithm 2.21 revisited) Routine to interpret change

of style commands
Argument OP indicates the desired style.
Constant FIRST-FILL-OP first fill op = —16
BEGIN
IF OP < FIRST-FILL-OP; THEN decode op and set polygon fill style
ELSE decode op and set line style;
RETURN;
END;

The INTERPRET algorithm must also be extended to handle polygon com-

mands. When a polygon command is discovered, control is then transferred to the
DOPOLYGON command, which processes the polygon.

3.6 Algorithm INTERPRET(START, COUNT) (Algorithm 2.22 revisited) Scan the

display file performing the instructions
Arguments START the starting index of the display-file scan
COUNT the number of instructions to be interpreted
Local NTH the display-file index
OP, X, Y the display-file instruction
BEGIN
a loop to do all desired instructions
FOR NTH = START TO START + COUNT — 1 DO
BEGIN
GET-POINT(NTH, OP, X, Y);
IF OP < -31 THEN DOCHAR(OP, X, Y)
ELSE IF OP < 1 THEN DOSTYLE(OP)
ELSE IF OP = 1 THEN DOMOVE(X, Y)
ELSE IF OP = 2 THEN DOLINE(X, Y)
ELSE DOPOLYGON(OP, X, Y, NTH);
END;
RETURN;
END;

The polygon command indicates that there is a polygon and indicates how many
sides it has. If the user has requested filled polygons by setting the SOLID flag, then
this information is given to the FILL-POLYGON routine which will do the actual fill-
ing in. In all other respects, the polygon command is treated as a move.
 potycons 79

3.7 Algorithm DOPOLYGON(OP, X, Y, INDEX) Routine to process a polygon com-

mand
Arguments OP XY the display-file instruction
INDEX the position in the display file of the instruction
Global SOLID aflag to indicate if the polygon should be filled in
BEGIN
IF SOLID THEN FILL-POLYGON(INDEX);
DOMOVE(X, Y);
RETURN;
END;

FILLING POLYGONS
One way of filling polygons is to first draw the edges of the polygon in a blank frame
buffer. Then starting with some ‘‘seed’’ point known to be inside the polygon, we set
the intensity to the interior style and examine the neighboring pixels. We continue to
set the pixel values in an increasing area until we encounter the boundary pixels. This
method is called a flood fill because color flows from the seed pixel until reaching the
polygon boundary, like water flooding the interior of a container. The flood-fill method
can be quite useful in some cases. It will work with any closed shape in the frame buf-
fer, no matter how that shape originated. However, it requires a frame buffer free of
pixels with the polygon interior style, in order to avoid confusion. It also requires a
seed pixel.
We could draw solid polygons by considering every pixel on the screen, applying
our inside test, and setting those pixels which satisfied it. This would avoid the need
for a seed pixel, but the method would be rather costly. Many pixels can be im-
mediately eliminated by comparing them with the maximum and minimum boundary
points. We really need to consider only those pixels which lie within the smallest rec-
tangle which contains the polygon. If we first find the largest and smallest y values of
the polygon, we need to consider only points which lie between them. Let us suppose
that we start with the largest y value and work our way down, scanning from left to
right as we go, in the manner of a raster display. Our constructed test lines will be the
horizontal lines at the current y scanning value. Many problems in computer graphics
can be approached a scan line at a time. Algorithms which take this approach are called
scan-line algorithms. Often we can take advantage of what we learned in processing one
scan line to make the calculation easier for the next scan line. (See Figure 3-11.)
We could draw the boundary of the polygon in a blank frame buffer and then
examine the pixels in the box around the polygon, scan line by scan line. Moving
across the scan line, when we encounter a pixel with the intensity of the boundary, we
enter the polygon. Subsequent pixels are given the interior intensity until we encounter
a second boundary pixel. The problems with this scheme are that we must start with a
frame buffer free of pixels with the polygon boundary intensity and we must be careful
about cases where two polygon edges meet at a single pixel. We can avoid these prob-
lems by determining the polygon boundary values directly from the polygon instruc-
tion, instead of from the frame buffer. Using the display-file instructions, we can deter-
80 CHAPTER THREE

A
Ze
Awe ea))
REE
Ze HEEeNTE
i ie EN
oS ee a
NR ee ee)
\cheese
a sn gen

FIGURE 3-11
Filling along scan lines.

mine where the scan line crosses the polygon boundary. To test a point, we do not have
to compute the intersection of our scan line with every polygon side. We need to con-
sider only the polygon sides with endpoints straddling the test line. (See Figure 3-12.)
It will be easier to identify which polygon sides should be tested if we first sort
the sides in order of their maximum y value. As we scan down the polygon, the order
in which the sides must be considered will match the order in which they are stored.
Each time we step down to a lower y value, we can examine the sides being con-
sidered, in order to determine whether we have passed their lower endpoints. If we
have stepped past the lowest y value of a side, it may be removed from the set of sides
being considered. Now for each y value we know exactly which polygon sides can be
crossed. (See Figure 3-13.)
We maintain our list of sides so that all the sides which are currently being
considered will be grouped together. We shall keep two pointers to mark the bound-
aries of the group, START-EDGE and END-EDGE. All edges stored with list indices
greater or equal to START-EDGE and less than END-EDGE should be considered. An
edge in the list before START-EDGE has yet to be encountered. Those which lie be-
hind END-EDGE will have been passed.

Consider only these sides

Scan line

FIGURE 3-12
Consider only the sides which intersect the scan line.
 PoLycons 8]

Remove this side

from consideration

Consider
this side
FIGURE 3-13
When the scan line passes the bottom endpoint of
a side, remove it from consideration. When the
scan line passes the top endpoint of a side, con-
sider the side.

Our task becomes setting those pixels on the horizontal scan line which lie inside
the polygon. It is really not necessary to examine every pixel on the horizontal line.
We can think of the polygon as breaking up a horizontal scan into pieces. (See Figure
3-14.) According to the even-odd definition of the polygon interior, the pieces alternate
light and dark. If we know the endpoints of the pieces—that is, the points where the
scan line crosses the polygon’s sides—then we can use our vector generator (or equiva-
lent) to fill in the entire piece. We do not have to consider each pixel in the scan-line
segment individually. Suppose we compute the x values for all intersections of poly-
gon sides with a given horizontal line, and then sort these x values. The smallest x
value will be the left polygon boundary. At this point the polygon begins. The next x
value indicates where the polygon ends. Therefore, a line segment drawn between
these values will fill in this portion of the polygon. We can pair up the sorted x values
in this manner for passage to our line-drawing routine. (See Figure 3-15.)
In summary, an algorithm for filling a solid polygon should begin by ordering
the polygon sides on the largest y value. It should begin with the largest y value and
scan down the polygon. For each y, it should determine which sides can be intersected
and find the x values of these intersection points. The x values are sorted, paired, and
passed to a line-drawing routine.

FIGURE 3-14
The polygon breaks the scan line into pieces.
82 CHAPTER THREE

FIGURE 3-15
x values are paired and used for line drawing.

The algorithm which performs the yx scan and fills in the polygon is called
FILL-POLYGON. It begins by retrieving the polygon shape information from the dis-
play file and sorting it by largest y value. This is accomplished by means of the
LOAD-POLYGON algorithm. The filling in of the polygon is done by repeating five
steps. The first step is to determine if any additional polygon sides should be consid-
ered for this scan line. The INCLUDE routine makes this determination. The second
step is to sort the x coordinates of the points where the polygon sides cross the scan
line so that they may be easily paired. This is done by the XSORT routine. The third
step is to actually turn on the pixels between the polygon edges, which is done by
FILL-SCAN. Next, the current scan line is decremented; and finally, the
UPDATE-X-VALUES routine determines the points of intersection of the polygon with
this new scan line, and removes from consideration any edges which have been pass-
ed. These steps are repeated until all polygon edges have been passed by the scanning
process.
3.8 Algorithm FILL-POLYGON(INDEX) Fills in a polygon
Arguments INDEX the display-file index of the instruction
Global YMAX an array of upper y coordinates for polygon sides
SCAN-DECREMENT the size of a scan-line decrement
Local EDGES the number of polygon sides considered
SCAN the y value of the scan line
START-EDGE, END-EDGE indicate which polygon sides are crossed by the
scan line
BEGIN
load global arrays with the polygon vertex information
LOAD-POLYGON(INDEX, EDGES);
are there enough sides to consider
IF EDGES < 2 THEN RETURN;
set scan line
SCAN <— YMAX[1];
initialize starting and ending index values for sides considered
START-EDGE < 1:
END-EDGE <1;
 PoLycons 83

fill in polygon
pick up any new sides to be included in this scan
INCLUDE(END-EDGE, EDGES, SCAN);
repeat the filling until all sides have been passed
WHILE END-EDGE # START-EDGE; DO
BEGIN
make sure the x values are in order
XSORT(START-EDGE, END-EDGE -— 1);
fill in the scan line
FILL-SCAN(END-EDGE, START-EDGE, SCAN);
next scan line
SCAN < SCAN — SCAN-DECREMENT;
revise x values
UPDATE-X-VALUES(END-EDGE — 1, START-EDGE, SCAN);
and see if any new edges should be considered
INCLUDE(END-EDGE, EDGES, SCAN);
END;
RETURN;
END;

Now let’s consider in more detail what we would like to know about each poly-
gon edge. We would like to know the largest and smallest y-coordinate values. The
largest y value indicates at which point in the scanning process to include this edge.
The smallest y value will determine when the line has been passed and need no longer
be considered. We shall also need to store x-coordinate information so that we can de-
termine where the edge will intersect the scan line. For this purpose, the x value of the
endpoint with largest y value should be saved. Thus we save both x and y coordinates
of the endpoint which will first be encountered in the scanning process. Now as we
step down through successive scan lines, the point of intersection will shift in x. The
amount by which it shifts can be determined if we know the change in x for a change
in y along the edge, that is, if we know the reciprocal of the slope of the edge. These
four items (X-TOP, Y-TOP, YBOTTOM, INVERSE-SLOPE) are all that we need to
know about each side. (See Figure 3-16.)
“We shall want to store this information for all nonhorizontal lines. Horizontal
lines can be ignored because they are parallel to the scan lines and cannot intersect
them. We shall want to store this information ordered according to Y-TOP, because this
is the order in which it will be accessed. The retrieval and storing of the edge informa-

(X-TOP, Y-TOP)

JANVERSE SLOPE

FIGURE 3-16
— Y-BOTTOM Parameters stored for each side of the polygon.
84 CHAPTER THREE

tion are performed by a routine which we have called LOAD-POLYGON. The al-
gorithm begins by setting one endpoint of the edge equal to the vertex in the original
polygon command. The point is converted from normalized coordinates to actual de-
vices coordinates. We add 0.5 to get the effect of rounding the point to a pixel number.
The y value is rounded right away by the INT function, which is the same as FLOOR
for positive arguments. The x value will be rounded at every scan line.
The LOAD-POLYGON algorithm next steps through the display file. It retrieves
a new vertex by means of the GET-POINT routine. This vertex becomes the second
endpoint of a polygon edge (the first endpoint came from the previous step). Horizon-
tal edges are ignored, but information for nonhorizontal lines is saved in order by the
POLY-INSERT routine. When all edges have been considered, the algorithm returns
the number of edges actually saved.

3.9 Algorithm LOAD-POLYGON(I, EDGES) Aroutine to retrieve polygon side

information from the display file. Positions are converted to actual screen coordinates
Arguments I the display-file index of the instruction
EDGES for return of the number of sides stored
Global WIDTH-START, HEIGHT-START starting index of the screen
WIDTH width in pixels of the screen
HEIGHT height in pixels of the screen
Local X1, Y1, X2, Y2 edge endpoints in actual device coordinates
Il for stepping through the display file
K for stepping through the polygon sides
DUMMY for a dummy argument
SIDES the number of sides on the polygon
BEGIN
set starting point for a side
GET-POINT(I, SIDES, X1, Y1);
X1<— X1 * WIDTH + WIDTH-START + 0.5;
adjust y coordinate to nearest scan line
Y1 <—INT(Y1 * HEIGHT + HEIGHT-START + 0.5);
get index of first side command
<a
initialize an index for storing side data
EDGES < 1;
a loop to get information about each side
FOR K = 1 TO SIDES DO
BEGIN
get next vertex
GET-POINT(I1, DUMMY, X2, Y2);
X2 <— X2 x WIDTH + WIDTH-START + 0.5;
Y2 <— INT( Y2 * HEIGHT + HEIGHT-START + 0.5):
see if horizontal line
IFY1 = Y2 THEN X1 < X2
ELSE
BEGIN
save data about side in order of largest y
POLY-INSERT(EDGES, X1, Y1, X2, Y2);
 potycons §85

increment index for side data storage

EDGES <— EDGES + 1;
old point is reset
Yoi<aye:
XI xX2;
END;
<a
END;
set EDGES to be a count of the edges stored.
EDGES < EDGES -— 1;
RETURN;
END;

The POLY-INSERT algorithm is basically an insertion sort. It determines the

maximum y value for the two endpoints and compares it with previously entered edges
to determine where in the sequence the new edge belongs. It begins with the last ele-
ment, to see if it belongs at this end. If the new edge’s maximum yvalue is smaller
than that of all previous edges, it is entered at this end. If not, the last edge is moved
down one position, opening a possible storage location in the next-to-the-last place.
This comparison with, and shifting of, the edges is continued until the appropriate po-
sition for the new edge is found, at which point information for the new edge is in-
serted. The data is stored in four separate arrays (one for each type of information).
These arrays should be dimensioned to match the maximum number of sides a polygon
can have in the system (one entry for each side). We have named the arrays YMAX to
save Y-TOP, YMIN to save Y-BOTTOM, XA to save X-TOP, and DX to hold the
change in x for each scan decrement. For a constant scan decrement, there will be a
fixed change in the x-intersection value for each scan. We can find the amount that the
intersection point will shift by multiplying the rate of change in x for a change in y by
the scan decrement. The scan decrement is the distance between the scan lines actually
being filled. Usually this will be 1, so that every scan line is filled, but by allowing
other values, we can achieve different fill styles. For example, a scan decrement of 2
would fill every other scan line. (See Figure 3-17.) We use the INT function to round
the x values to integer pixel positions in determining XA and DX. This is so the sides
of the polygon will match the boundary drawn with the Bresenham algorithm.

3.10 Algorithm POLY-INSERT (J, X1, Y1, X2, Y2) The ordered insertion of polygon
edge information
Arguments J insertion index
X1, Y1, X2, Y2 endpoints of the polygon side (y values rounded)
Global YMAX, YMIN, XA, DX arrays for storage of polygon edge information
SCAN-DECREMENT step between filled scan lines
Local J1 for stepping through the stored edges
YM the maximum y value of the new edge
BEGIN
insertion sort into global arrays on maximum y
Jy;
find the largest y
 86 CHAPTER THREE

FIGURE 3-17
Using scan decrement for fill styles.

YM < MAX(Y1, Y2);

find correct insertion point, moving items out of the way as we go
WHILE J1 ¥ 1 AND YMAX(J1 — 1) < YM DO
BEGIN
move up the insertion slot
YMAX{[J1] — YMAX{[J1 — 1];
YMIN[J1] — YMIN[J1 — 1];
XA[J1] — XA[J1 — 1);
DX[J1] — DX[J1 — 1);
Ue eee
END;
insert information about side
YMAX{[J1] — YM;
DX[J1] <— (INT(X2) — INT(X1)) /(¥2 — Y1)) *( — SCAN-DECREMENT);
see which end is on top
IFY1 > Y2 THEN
BEGIN
YMIN[J1] — Y2;
XA[J1] — INT(X1);
END
ELSE
BEGIN
YMIN{[J1] —Y1;
XA[J1] <— INT(X2);
END;
RETURN;
END;

The edge information for the polygon has been

stored for our use, but we need
not consider every edge with every scan line. We
need to maintain the START-EDGE
and END-EDGE pointers to delimit the edges
of interest. (See Figure 3-18.)
 PoLycons 87

YMAX XA DX YMIN

Set of edges — START-EDGE

crossed by | |
scan line —END-EDGE

FIGURE 3-18
Table entries for edges which cross
Table of edges the current scan line.

The algorithm we have called INCLUDE adds new edges to the group being con-
sidered. Because of the order in which we have stored the edges, the next edge to be
included will be the next edge in the array. To include the new edge, we only have to
increment the END-EDGE boundary. For each new scan line, the INCLUDE al-
gorithm checks the largest y value for the next edge; if the scan has gone below this
value, then END-EDGE is incremented to include the edge.

3.11 Algorithm INCLUDE(END-EDGE, FINAL-EDGE, SCAN) Include any edges

* newly intersected by the scan line
Arguments END-EDGE index of the side being considered for inclusion
FINAL-EDGE index of last side
SCAN position of current scan line
Global YMAX, XA, DX arrays of edge information
SCAN-DECREMENT the size of a scan-line decrement
BEGIN
WHILE END-EDGE = FINAL-EDGE AND YMAX[END-EDGE] = SCAN DO
r include a new edge
END-EDGE < EDGE-END + 1;
RETURN;
END;

The edge information between START-EDGE and END-EDGE is kept ordered

on the x-intersection value. (See Figure 3-19.)
The task of maintaining this ordering belongs to the algorithm called XSORT.
The XSORT routine steps through the currently active edges. If the position is correct,
then nothing happens. If, however, the element is out of place, it is ““bubbled up”’ to
its correct position by a series of exchanges with its neighbors. When a new edge is en-
tered, it may have to be shuffled down the array to place it in order. In subsequent
checks, however, the edges will almost always be in their correct positions. The excep-
tion is when polygon sides are allowed to cross.
88 CHAPTER THREE

START-EDGE ;
Scan line

END-EDGE

FIGURE 3-19
Table entries sorted on x.

3.122 Algorithm XSORT(START-EDGE, LAST-EDGE) Checking the order of the x

intersection
Arguments START-EDGE index of the first of the edges considered
LAST-EDGE index of the last edge whose order is to be checked
Global YMIN, XA, DX arrays of edge information
Local K, L for stepping through the edges
T temporary storage for the exchange
BEGIN
FOR K = START-EDGE TO LAST-EDGE DO
BEGIN
L<K;
WHILE L > START-EDGE AND XA[L] < XA[L — 1] DO
BEGIN
T<—YMIN[L];
YMIN[L] <— YMIN[L — 1];
YMIN[L — 1] <T;
T< XA[L];
XA[L] <— XA[L — 1];
XA(L — 1] <—T
T <— DXIL];
- DX[L] — DX[L - 1);
DX[L — 1]<T;
L<L-l1;
END;
END;
RETURN;
END:

The next algorithm, FILL-SCAN, actually fills in a scan line. It contains a loop
which steps through all current intersection points, connecting pairs with line seg-
ments. The actual line-segment drawing is done by FILLIN.
 potycons 89

3.13 Algorithm FILL-SCAN(END-EDGE, START-EDGE, SCAN) Fill in the scan

line
Argument START-EDGE, END-EDGE indicates which edges are crossed by the scan
line
SCAN the position of the scan line
Global XA an array of edge intersection positions
Local NX the number of line segments to be drawn
J for stepping through the edges
K for stepping through line segments
BEGIN
NX < (END-EDGE — START-EDGE) / 2;
J <— START-EDGE;
FOR K = 1TONX DO
BEGIN
FILLIN(XA[J], XA[J + 1], SCAN);
Ye See
END;
RETURN;
END;

The FILLIN routine may depend upon the type of display system being used. If
FILLIN uses some vector-generating routine provided by the display, FILLIN may
have to store the current line style, set the line style to the polygon interior style, move
to one of the endpoints, draw a line to the other endpoint, and reset the line style and
pen position to their original values. Alternatively, we can write a simple FILLIN al-
gorithm based on an assumed frame buffer (or some other method of setting individual
pixels), as were the vector generation algorithms of Chapter 1.

3.14 Algorithm FILLIN(X1, X2, Y) Fills in scan line Y from X1 to X2

Arguments X1, X2 end positions of the scan line to be filled
Y the scan line to be filled
Global FILLCHR intensity value to be used for the polygon
7 FRAME the two-dimensional frame buffer array
Local X for stepping across the scan line
BEGIN
IF X1 = X2 THEN RETURN;
FOR X = X1 TO X2 DO FRAME[X, Y] <— FILLCHR;
RETURN;
END;

For each new scan line, we must examine the currently active edges to see if ony
have been passed. If the edge should still be considered, the intersection value for the
new scan line should be calculated; if it has been passed, the edge should be removed
from consideration. This is the job of the algorithm called UPDATE-X-VALUES. To
determine if an edge still crosses the scan line, the lowest y value is examined. To up-
 90 CHAPTER THREE

date the intersection point, the constant step size (determined in INCLUDE) is added.
To remove an edge from consideration, the information for previous edges is moved up
one position in the arrays. This overwrites the deleted edge and allows incrementing of
the START-EDGE boundary. (See Figure 3-20.)

3.15 Algorithm UPDATE-X-VALUES(LAST-EDGE, START-EDGE, SCAN) Update

points of intersection between edges and the scan line
Arguments START-EDGE and LAST-EDGE limits of current edge list
SCAN the current scan line
Global XA, DX, YMIN arrays of edge information
Local K1 index of edge being considered for update
K2 index of where to store the updated edge
BEGIN
K2 <— LAST-EDGE;
FOR K1 = LAST-EDGE TO START-EDGE DO
BEGIN
check each edge
IF YMIN[K1] < SCAN THEN
BEGIN
the edge is still active so update its x values
XA[K2] — XA[K1] + DX[K1];
IF K1 ~ K2 THEN
BEGIN
YMIN[K2] <— YMIN[K1];
DX[K2] — DX[K1];
END;
decrement K2 so the edge won’t get overwritten
Ka Ke
END;
END;
START-EDGE < K2 + 1;
RETURN;
END;

START-EDGE at |
START-EDGE

Entry to be =
removed ee

END-EDGE --— END-EDGE

Before After
FIGURE 3-20
Removal ofa table entry.
 potycons 9]

The above algorithms allow us to fill in the interiors of polygons efficiently. We

should note, however, that if we are not careful, the method used for determining the
polygon boundary may be different from that used by the line generation algorithm to
outline the polygon. The interior may turn out to be a pixel wider than the edge, which
can be apparent at low resolution.

FILLING WITH A PATTERN

We mentioned that polygon fill styles might be patterns. Patterns are most easily im-
plemented on raster display devices. A pattern is a grid of pixel values which is repli-
cated like tiles to cover the polygon area. A pattern is often fixed, or registered, to the
imaging surface so that if two polygons are filled with the pattern and placed side by
side, the pattern will match at the boundary. We can imagine taking the pattern and rep-
licating it to cover the entire imaging surface, and then erasing it anywhere outside of
the polygons which use it. (See Figure 3-21.)
Assuming a frame buffer (or some other means of setting individual points), we
can show how patterns might be added to our graphics system. We can set up a table of
patterns and use the fill style to select one. We shall provide a routine for placing pat-
terns into the table. We shall also give a version of FILLIN which uses the table. Some
examples of 4 x 4 patterns are shown in Figure 3-22.
The pattern table can be composed of three arrays—PATTERN-X, PATTERN-Y,
and PAFTERNS— where PATTERN-X and PATTERN-Y are arrays of numbers which

A pattern

Replicated across the display

FIGURE 3-21
Shown only within polygon boundaries Replication of a pattern.
92 CHAPTER THREE

FIGURE 3-22
Some 4 X 4 patterns.

specify the size of each pattern, and PATTERNS is an array of two-dimensional arrays
which are the actual patterns. (One way to implement PATTERNS is to make it a three-
dimensional array, but this may not be as flexible as alternatives offered by some lan-
guages.) We place an individual pattern into the pattern table with the following al-
gorithm.

3.16 Algorithm SET-PATTERN-REPRESENTATION(PATTERN-INDEX, PAT-X,

PAT-Y, NEW-PATTERN) Enters a pattern into the pattern table
Arguments PATTERN-INDEX the place in the pattern table to save the pattern
PAT-X, PAT-Y the dimensions of the pattern
NEW-PATTERN the pattern, a two-dimensional array of intensity values
Global PATTERN-X, PATTERN-Y, PATTERNS the pattern table
BEGIN
PATTERN-X[PATTERN-INDEX] < PAT-X;
PATTERN-Y[PATTERN-INDEX] < PAT-Y;
the following statement depends upon the implementation of PATTERNS, and may
entail the copying of all the individual intensity values
PATTERNS[PATTERN-INDEX] <- NEW-PATTERN;
RETURN;
END;

The FILLIN algorithm is revised as follows:

3.17 Algorithm FILLIN(X1, X2, Y) (Revision of algorithm 3.14) Fills in scan line Y
from X1 to X2
Arguments X1, X2 end positions of the scan line to be filled
Y the scan line to be filled
Global FILL-PATTERN pattern table index of the pattern to use
PATTERN-X, PATTERN-Y, PATTERNS the pattern table
FRAME the two-dimensional frame buffer array
 POLYGONS 93

Local X for stepping across the scan line

PX, PY for accessing the pattern
PATTERN-TO-USE the pattern to be used in filling
PAT-X, PAT-Y the x and y dimensions of the pattern
BEGIN
IF X1 = X2 THEN RETURN;
PAT-X < PATTERN-X[FILL-PATTERN];
PAT-Y <— PATTERN-Y[FILL-PATTERN];
if a local array is used as shown in the following statement,
then the implementation should avoid copying of individual elements
PATTERN-TO-USE < PATTERNS[FILL-PATTERN];
PX < MOD(X1, PAT-X) + 1;
PY — MOD(Y, PAT-Y) + 1;
FOR X = X1 TO X2 DO
BEGIN
FRAME[X, Y] <— PATTERN-TO-USE[PX, PY];
IF PX = PAT-X THEN PX < 1
BESE PX PX 1s
END;
RETURN;
END;

The MOD functions give the remainder after division of the first element by the
second: This is what causes the replication of the pattern. If we have a value X1 or Y
which is larger than the pattern size, the MOD function wraps PX and PY back into
values within the pattern, thereby repeating the pattern. The Is are added into this cal-
culation because we have assumed our arrays are dimensioned with starting index 1.
We set the local variables PAT-X, PAT-Y, and PATTERN-TO-USE in order to remove
the address computation from the loop. Actually, it would be even better to make these
variables global and to set them in DOSTYLE (the only reason we did not do so is that
DOSTYLE is too system-dependent for a clean example).
On many devices there are only two pixel states, on or off. For these displays,
the frame buffers and pattern tables can be compactly implemented by using individual
bits to describe the pixel states. Pattern dimensions can be chosen to lie on word
boundaries, and the FILLIN algorithm can be made more efficient by dealing with en-
tire words of pixel values.

INITIALIZATION
To finish this chapter’s algorithms, an initialization routine is needed to set default
values for filling and fill-style parameters. If filling with patterns is possible, then the
pattern table should be initialized to a default set of patterns.

3.18 Algorithm INITIALIZE-3 Routine to initialize the system

Local P for stepping through the pattern table
Constants _MINIMUM-FILL-OP opcode for first fill style = — 16
NUMBER-OF-PATTERNS the size of the pattern table
94 CHAPTER THREE

DEFAULT-PAT-X, DEFAULT-PAT-Y size of the default pattern

DEFAULT-PATTERN a default pattern array
BEGIN “
INITIALIZE-2;
DOSTYLE(MINIMUM-FILL-OP);
SET-FILL(FALSE);
the following loop is included only if polygons may be filled with patterns
FOR P = 1 TO NUMBER-OF-PATTERNS DO
SET-PATTERN-REPRESENTATION(P, DEFAULT-PAT-X, DEFAULT-PAT-Y,
DEFAULT-PATTERN);
RETURN;
END;

ANTIALIASING
In Chapter 1 we mentioned a technique for smoothing the jagged steps in a line which
is introduced by the quantization to finite-sized pixels. Aliasing is a problem for the
edges of polygons just as it is for lines, and several antialiasing techniques which use
the shading of gray-level displays to reduce the effects have been developed. One
technique is to calculate what fraction of a pixel is actually covered by the polygon and
how much is background. The pixel intensity displayed would then be the average of
the background and polygon intensities weighted by their relative areas.
Another approach is to generate the scene at a higher resolution than that which
will actually be used, and then to average the intensity values of neighboring pixels to
determine the intensity to be displayed. Increasing the resolution between four and
eight times gives good results. Note that this need not be as much work as it seems.
The antialiasing can be incorporated as part of the polygon-filling algorithm. Antialias-
ing need only be applied to the points on the edge (interior points will have the full
polygon intensity). It can be carried out scan line by scan line, so we don’t need exces-
sive amounts of memory.

AN APPLICATION
Before leaving this chapter, let’s discuss how we might use some of the routines which
we have developed. Let’s consider how we might write a package to draw a bar graph
of some data. We have already seen in Chapter 2 how the axes might be drawn and
labeled, so we shall concentrate on how to generate the bars. We can make a bar out of
a four-sided polygon. (See Figure 3-23.)
Suppose that we construct such a polygon with lower-left corner at the current
pen position by using the POLYGON-REL-2(AX, AY, 4) command. Then the neces-
sary AX and AY array values would be

AX = 0, DX, 0, —DX
AY = 0, 0, DY, 0
 potycons 95

dy

FIGURE 3-23
dx Constructing a solid bar from a polygon.

Here, DX will give the width of the polygon and DY will give its height. Now we can
fix DX at the width which we want our bar graph’s bars to have. All bars should have
the same width, but the height of each bar depends upon the input data. Each bar
might have a different height. We can see, however, that it is easy to change the bar
height; we need to change only the value of the DY parameter in the AY array. Thus by
poking the appropriate value into the AY array and calling the POLYGON-REL-2
routine, we can generate a bar with any height we choose. Now what we want to do is
to drawa series of bars, each at a different horizontal position and each with a height
representing the data value for that position. This is easy because we have used arela-
tive polygon command. To position the bar, we need only move the pen to the correct
starting position before drawing the polygon. (See Figure 3-24.)

MOVE-ABS-2(X, Y);
POLYGON-REL-2(AX, AY, 4);

Now we can place these instructions within a loop. Each time through the loop,
we get one of the data values to be plotted. We increment Xin order to position the pen
at the starting point for the next bar. We calculate the height of the bar from the data
value and put it into the AY array. And we call the MOVE and POLYGON algorithms
to actually draw the bar. Each time through the loop, another bar would be drawn until
the graph is complete. (See Figure 3-25.)

=3 FIGURE 3-24
Boe (XN) A MOVE command will position a relative polygon.
96 CHAPTER THREE

FIGURE 3-25
A bar graph constructed by repeated use of the POLYGON
command.

We might note that if the bars are equally spaced and drawn sequentially, the
explicit MOVE-ABS-2 command can be replaced by the implicit relative move done
by the POLYGON-REL-2 algorithm.

FURTHER READING
A discussion of winding numbers and inside tests may be found in [NEW80]. We rep-
resented polygons as a list of vertex coordinates; alternative representations are pre-
sented in [BUR77] and [FRA83]. To fill a polygon, we broke it down into individual
scan lines. This technique is called scan conversion and is discussed in [BAR73] and
([BAR74]. An alternative approach to filling regions such as polygons is to draw their
border in the frame buffer and then, starting with a seed pixel within the polygon,
progressively examine each pixel, changing its color to the interior style until the
boundary is encountered. This technique is called flood fill or seed fill and is discussed
in [LIE78]. Instead of filling from a single seed, we can scan the area to be filled,
building regions. When two regions merge, we note their equivalence; then on a sec-
ond pass we shade all regions associated with the interior [DIS82]. The decomposition
of polygons into trapezoids or triangles is described in [FOU84], [GAR78], [JAC80],
and [LIT79]. Polygons can also be described as the differences of convex polygons
[TOR84]. An attempt to save the scan-converted polygons in the display file rather
than in a frame buffer is described in [SPR75]. Filling polygons on calligraphic de-
vices with patterns of lines is discussed in [BRA79]. Polygon filling is also discussed
in [AGK81], [DUN83], [PAV78], [PAV81], [POL86], and [SHA80]. A discussion of
antialiasing of polygons is given in [BAR79] and [CRO81]. Our polygon primitives
are based on proposed extensions to the CORE system, which are described in
[GSPC79]. ;

[AGK81] Agkland, B. D., ‘‘The Edge Flag Algorithm—A Fill Method for Raster Scan Displays,’’ JEEE
Transactions on Computers, vol. C-30, no. 1, pp. 41-47 (1981).
([BAR73] Barrett, R. C., and Jordan, B. W., Jr., ‘‘A Scan Conversion Algorithm with Reduced Storage Re-
quirements,’’ Communications of the ACM, vol 16, no. 11, pp. 676-682 (1973).
{[BAR74] Barrett, R. C., and Jordan, B. W., Jr., ‘‘Scan-Conversion Algorithms for a Cell Organized Raster
Display,’’ Communications of the ACM, vol. 17, no. 3, pp. 157-163 (1974).
[BAR79] Barros, J., Fuchs, H., ‘‘Generating Smooth 2-D Monocolor Line Drawings on Video Displays,’’
Computer Graphics, vol. 13, no. 2, pp. 260-269 (1979).
 POLYGONS 97

[BRA79] Brassel, K. E., Fegeas, R., ‘‘An Algorithm for Shading of Regions on Vector Display Devices,”
Computer Graphics, vol. 13, no. 2, pp. 126-133 (1979).
[BUR77] Burton, W., ‘‘Representation of Many-Sided Polygons and Polygonal Lines for Rapid Process-
ing,’’ Communications of the ACM, vol. 20, no. 3 , pp. 166-171 (1977).
[CRO81] Crow, F. C., ‘‘A Comparison of Antialiasing Techniques,’ JEEE Computer Graphics and Applica-
‘tions, vol. 1, no. 1, pp. 40-48 (1981).
[DIS82] Distante, A., Veneziani, N., ‘‘A Two-Pass Filling Algorithm for Raster Graphics,’’ Computer
Graphics and Image Processing, vol. 20, no. 3, pp. 288-295 (1982).
[DUN83] Dunlavey, M. R., “‘Efficient Polygon-Filling Algorithms for Raster Displays,’ ACM Transactions
on Graphics, vol. 2, no. 4, pp. 264-273 (1983).
[FOU84] Fourner, A., Montuno, D. Y., ‘‘Triangulating Simple Polygons and Equivalent Problems,’” ACM
Transactions on Graphics, vol. 3, no. 2, pp. 153-174 (1984).
[FRA83] Franklin, W. R., ‘“Rays—New Representation for Polygons and Polyhedra,’’ Computer Vision,
Graphics, and Image Processing, vol. 22, no. 3, pp. 327-338 (1983).
[GAR78] Garey, M. R., Johnson, D. S., Preparata, F. P., Tarjan, R. E., ‘‘Triangulating a Simple Polygon,”’
Information Processing Letters, vol. 7, no. 4, pp. 175-179 (1978).
[GSPC79] Graphic Standards Planning Committee, ‘‘Status Report Part III: Raster Extensions to the CORE
System,’” Computer Graphics, vol. 3, no. 3, pp. II-1-III-39 (1979).
[JAC80] Jackson, J. H., ‘‘Dynamic Scan-Converted Images with a Frame Buffer Display Device,’’ Com-
puter Graphics, vol. 14, no. 3, pp. 163-169 (1980).
[LIE78] Lieberman, H., ‘‘How to Color in a Coloring Book,’’ Computer Graphics, vol. 12, no. 3, pp. 111-
116 (1978).
[LIT79] Little, W. D., and Heuft, R., ‘‘An Area Shading Graphics Display System,’’ JEEE Transactions on
Computers, vol. c-28, no. 7, pp. 528-531 (1979).
[NEW80] Newell, M. E., and Sequin, C. H., ‘‘The Inside Story on Self-Intersecting Polygons,’’ Lambda,
vol. 1, no. 2, pp. 20-24 (1980).
[PAV78] Pavlidis, T., ‘‘Filling Algorithms for Raster Graphics,’’ Computer Graphics, vol. 12, no. 3, pp.
161-166 (1978).
[PAV81] Pavlidis, T., ‘‘Contour Filling in Raster Graphics,’’ Computer Graphics, vol. 15, no. 3, pp. 29-36
(1981).
[POL86] Polik, W. F,, ‘‘Area Filling Algorithms,’’ Computer Language, vol. 3, no. 5, pp. 33-40 (1986).
[SHA80] Shani, U., ‘‘Filling Regions in Binary Raster Images: A Graph-Theoretic Approach,’’ Computer
Graphics, vol. 14, no. 3, pp. 321-327 (1980).
[SPR75] Sproull, R. F, Newman, W. M., ‘‘The Design of Gray-Scale Graphics Software,’’ Proceedings of
the Conference on Computer Graphics, Pattern Recognition, & Data Structure, pp. 18-20, IEEE Cat.
No. 75CH0981-1c, (1975).
[TOR84] Tor, S. B., Middleditch, A. E., ‘‘Convex Decomposition of Simple Polygons,’’ ACM Transactions
ron Graphics, vol. 3, no. 4, pp. 244-265 (1984).

EXERCISES
3-1 For each of the following polygons, indicate whether it is concave or convex:

(a) (b) (c)

 98 CHAPTER THREE

(d) (e) )

3-2 Sketch what would be drawn by a POLYGON-ABS-2(AX, AY, N) command

for each of the
following array pairs.
a) N=8
AXi=0:27-0.2, 01,0: 1)0.490.45 0:3; 0:3
AY = 0.1, 0.3, 0.3, 0.4, 0.4, 0.3, 0.3, 0.1
b) N=8
AX = 0,5,,0.5, 0.3, 0.3, 0.5,0:5--011, 0.1
AY ="071,'0.2, 0°2,0.3, 0.3.0.4, 0.4, 0.1
cj Ni= 8
A ="0:2 0.3. 0:2, 0:3, 0.2. Ol el 0ud
AY, =70.1,0-1,0.2)0.3, 0.45 14,0296.
Gd) Ne==5)
AX = 0.1, 0.1, 0.5, 0.9, 0.9
AY = 0.1, 0.6, 0.8, 0.6, 0.1
ie) Ne 27
AX = 0.5, 0.3, 0.4, 0.4, 0.6, 0.6, 0.7
AY’ ="0.7,0.5; 0.5,0-1, 0.1; 05205
3-3 Sketch what would be drawn by
MOVE-ABS-2(0, 0)
POLYGON-REL-2(AX, AY, N)
for each of the following array pairs:
a) N=4
AX =/0.1) 0517 O71; 2041
AY = 0.2, -0.1, 0.1, 0.1
b) N = 10
AX ='0.4,0, -0.2, 03, 0:1; 0.1/0.3, 6.2, 0, 0
AY'= "0.2, 072,022, 0, 0:2) 20 0120.0. 20 Go
cyeNe= 5
AX = 0:170.1;0.15 O:1,:011
AY = 0:20:8; <071; 0:0)-04
d) N= 10
AX = 0.5, -0.4, 0, 0.1, 0, 0.1, 0, 0.1, 0, 0.1
AY = 0:3, 0, -0.1, 0, 0:1, 0, 0:1, 0,-0.1. 0
e) N= 1)
AX = 0.1, 0.2, -0.1, 0.2, -0.1, 0.2, 0.2, —0.1, 0.2,
-0.1, 0.2
AY = 0.1; 0.2/0).0,2, 0, 012, 20 02 0, —0.2, 0,
-0.2
3-4 Give suitable values of argument arrays
for a POLYGON-ABS-2 command to draw each
the figures in Exercise 3-1. of
3-5 Give suitable values of argument arrays
for a POLYGON-REL-2 command to draw
the figures in Exercise 3-1, assuming that each of
the pen begins at the origin for each case.
al x
POLYGONS 99

3-6 For each of the polygons shown, indicate the interior regions
a) Using the even-odd method
b) Using the winding-number method -

a |
4+—--— «+——_!

PROGRAMMING PROBLEMS
3-1 Implement algorithms 3.1 through 3.15 and 3.18. Devise DOSTYLE and FILLIN routines
appropriate for your system.
3-2 Test the implementation of Programming Problem 3-1 by drawing a house as a five-sided
polygon. Show this house in outline and filled with at least two fill styles.
100 CHAPTER THREE

3-3 Write routines named

INQUIRE-FILL-STYLE(I)
and .

INQUIRE-FILLED(ON-OFF)
which return the current polygon interior style and the value of the SOLID flag.
3-4 Implement algorithms 3.16 and 3.17 (revised for your system if necessary) to allow filling
of polygons with patterns. Show a polygon filled with three different patterns.
3-5 Write a program to construct a picture out of N squares (each square should be a four-sided
0, 1 -
polygon). The vertices of the squares should be (J/N, 0) (1, J/N), (1 - GIN), 1), and
(J/N)), where J steps between 1 and N, inclusive.
a) Show the squares unfilled.
b) Show each square filled with a different interior style.
3-6 a) Write a program to draw a number of overlapping triangles on the display. The trian-
gles can have random positions, random shapes, and random interior styles.
b) Write a program to draw a number of overlapping random polygons. The polygons
should vary in the number of sides as well as in the position and interior style.
3-7 Consider the following map showing the four counties in the State of Invention.

(eilter:9)) eee (eS) (959)

GIEs:6)

(.9, .4)

Glee) (ea) CSE)

a) Construct this map out of polygons, using one polygon for each county. (Do not label
the vertices on your map.) re
b) Show your map with each county filled with a different interior style.
3-8 Consider the following table of widget sales and expenses.

Widget annual report

Month Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec

Sales 2000 4000 7000 6000 4000 8000 8000 7000 5000 6000 8000 9000
Expenses 5000 5000 4000 3000 2000 4000 5000 6000 8000 9000 6000 5000
 POLYGONS 101

a) Form a polygon out of the widget sales data and the lower-left and lower-right
corners
of the graph axis, and display this both unfilled and filled.

KK
x x
xk Ox
OK x

FAO OO IOOK * :
* we * *
% x * x*
* + x x
x xx: x x
*
x
x
k
*
x
x
*
*
x
x
x
*
*
*
* x
*x
x
x
x
x
*
x
*
*
x
x
*
x
x
*
x
x
x
RH
MM
HHH
MMMM
FEO OOOO OOOOOOO a GOGO GOGO OOOO GCG OOOO OOOROOISO OOOO OGRE OO OOO OGRA IK *

bk
KOK
KKK K
FORO OK
HOR AIOK AK
OOK OK TORR
ACH AK
EO ICO OOK SOON AEOK
EI CIOORO OO IK NOK SOO IO CICKAK
JOOS OOOO OOO IOKIK SOOO OOK OK
Petes tetetees ss ce gg FEO IOI IOK
*K OOOO OR ICO OK KOK OOOO OOK KK
2K FEO CO OOOIGOR AOKI IK JOO OOK EK
SOK FEO ORO OKI OK JOO OOOO OK 101K
JOO OOK FEO OOOO IK JEIOOI OOOO OOO IK
7 {EO IO OOK OOOO COO OOOO OIA OOOO OK OK OK KK
FAO IO IOK FEO OOOO OOO OOOO ICKIOK SEO OOOO OO IK
JARO IOIOK JEOOOO IOSOGOIOIO OOOO OK EOI OO A ORIOK KOKKOK
JOO OOOO IOK SEO OOOO OOOO IORI CIOK EOC OOOO OOIIOK OK
FESO OOOO OK SEO BIO OOOO OOOO OOOO GOO OOOO ORIG KOO OK OKOK
JOO OOOO OI IK JCC OOO OOOOOOO ODOR OOOO OOGOOOOOIS OOOO IK
SEO OOOO OOK OHO AOK FEO OOOO OOO OOOO OOOO OOO BOOK BORO OOOO OKI KOK
OOOO GOO OOOGIOIIOK a GOO GGORGGOOIOGERIOOOIOOIOIOOGOR GGG ORRIROIOK IOIIIA II
JAC OOO OOOO OOO OOOO SOOO SS GOGIOS OOOO ORIORO OOOO OOOO III OKA IK
OOOO OOOO OOOO COOOIKO OOOO GOCOOOOOOR DIOGO CICIOOOI OIAOODIOI OOO ARIO ORIG
JOG IOIOOOOOOOOOOOOOOR OOOO OOO OOOIOOOIOIOIG OSORIO OIGOIOOIDID ROO COOR CORO GO IOI IORI IOI
EOC COO GO OOOO OOOO OOO GC ROII ODIO OIOOODIOOOOR OTRO IGOR
JOO OO OOOO OOOO COOOOIOOOI OOOO OIOOO OOOO OOOO OIC OGIAGSIOIO OOOO CEI OOOGOO
EUS OOOO OOOO OOOOOIOO GOGO OROIOOGIOIOIR OIRO OOO ORR EGOOG RIOOOOOO IOI
YEO SICIOOOOOOIOOO RO GOOOOOOO GOO COORG OROIOIOOIO OREO ORGAO OBOIGOOOOOOIO OOOO AIA OK
OOOO GOO OOOO OOO GOO SOOOO GIOIOIOO OGG AOIOIOIOIOIOOIOOO ORIG IGO OKCOIOIIOOIOIOO OOO OK
JOO OSOOIOOIOOO OOO COCO OO OOO OOOO ORGS OOOO OIOIOODOOOO AEA RIOIOOIO ORO OOOO IORI
YOK OOOODIOIOOOOOGROO OOOO OOOOOOOOOCGGOOOIOOOOO OOOO OO OGOR OSI OOO OCA REO AGOGO ORR IOI
JOGO OOOO OOOO OOO OOOCIOOOOSOOOOOOOGOOOROIOIOIOI OOOO aC GRGGOO RAR IOC ACCOR AGC IRIORACOROIOKICICK
JES SIOOIDIOIODIOIOOOO GOR OOOOOOOOOODIOIOOIOO RG OODIOIOIOIOO OO ORO OOOO OROOO OOOO ORCI ARGO IOIOOKK
JAICOOOO OOOO OGG OOOOIOOORIGCOOIOIO OOO GORIOIOIOO OOOO RGIORIOIOIOIOROO ORI OOOO DORIA OIRO AAO OK
JEIOISIOO OOOO OGG ORO OOOO GOOIOIO OOOO GOI IOOO OOO HOI DIDIOIOIOIOIOI CORIO RIO IOI IOI AIK
JO IOIOIOIOOI OOO CGO OOO CORROOOO OOOO COI IOI OOO SCOOT ODI OIOOIOEIO OOOO] ODIO ICRA G OOOO OIC IK
JEOKIOOOOIOOOOOCO OOOO COCO OO GGG OO OOROO OOOO OOOO OOOO ORR OOO GO CRREC OOGOIOOCRIOOCO ORI K AIK
JE IOIOIOIO OOOO GOGOROOOOOGOOOIOOOOOOGOOIIOO GOOG AROIIOOIIIDIOIOIOIOIOIOIOOORIOKORIOIOIOIOIG GIDE SIA IK
EIOIOSOIOOIOOGOOI OOOO OOOO CORO OOOO IDIOIOI OOOO ABO ORIOCORI GO OOOO ORI ROO OOO IK
YETCIQOIISIOOOO OOOO COR OG OOOOO OOOO GOOCOOOO CORO OCUROCCO COORG GOO TOO OCI OCC IOR YOK IOOIOK
AEOOOIOOIDOOOIOIOOOOOOOROOROC OOOO CGOO OOO GOG OOOO OOOOIORI COOOOO IOC OO OOOO IIII K
YEOIOISIO OR OCI OOOO OR ICO GIOCZOKIOOI AGGOGROICIOOICO ROR ORAO GOO IEICE CCI IO OOK
FAOOISIOOI OOOO OOOO OOOO OOOO OGG IOC OCOO OIC GORIO OS AIORIORISIO SOG OBIOIOIOIOIOIA GEIACIICAGIAIIO AIK
JEG OIO ODOC COOSA COOOOIOOIOGOICI OOOO OOOO GOI CIORIOOIORIOIOIOGOIOOI OOOO OOOSIARIE RIGO ARIE AGGRO EI
102 CHAPTER THREE

b) Form a similar polygon for widget expense figures and plot both polygons unfilled.

tt ON 1k
+ + x
+ *k OK
++ + xk Ok
++ + 1k x
JOCK OK t + 2 s
x 4k + Ths
x x + + x Oe x
x 4K + +H *x
* Hk + + x
*
* x + x + x
* OK
x
x
* + *
* +k
* *
* + * * + x
* kK x
++ * ** a Se x
* * x
eRe x

** +x
++++++t+++ * * x i +
+ t+ ¢+
+
+ is
+ +
+ +
+ +
+ +
+ +
+ +
+ x ++ + +
+ +
+ +
+ + + -H
i* ++ +
+ +
+ +
+ +
+ +
+
+
' :+
+ +
+ +
+ EE EE TEE ttt eet tt td
TESEEEEEEEEEEEEI EEE EE EEE EEE EEE EE EEHE EEE

c) First, plot the sales polygon filled, and then superimpose the expense polygon un-
filled. Use an expense line style different from the sales interior style.
*xk +t
OK ae
OOK ++ +
++ OOK +
++ + OOK OK
OOOO OK + + FOO IK
FOO OOK KK + + SOO OK KK
FAO IOC + + OOOO
FAO OOK ORK OK + HOO OOK KOK
OOOO OKO OKKOK + SCHOO OOO
xx SOOO OOOO IGOR IOOK + JOHOR KOK K
OOK SEO OOOO CIR IOK ++ SOHO KK
OOK OK FORO OC HOK + AIO HAO ORKAK
OOK OOK AOK SEO CORK IOK 3K SOOO 9 OKKK OOK
2OOOOKOKOKK COO OOK COC ORK KKK JAK OOK KKK
FOO OOK OK XK FCO OOOO OOK OKOK K+ 7 KKK FOO + EOC KOK
FOO OOOO OOK SEO OOOO OKO OK OK FEO OOK + FO KK
OOOO OOK IOK OK CCK FOO OOOO OOK + ICO IO IK FRO OR KK OK OOK FAK RK
FAO OOO OK IOKKOK AK FEO OOOO IOI IOKEHO OR KOK K FOO OO OOO OOK OK OK
HET OOOO OK KOKAC AAO OOO CIC OIC ORO OSCE IOC OKO COOIOK ARGO IOKTH
7 SHO COOK KOIOK OOOO OOOO OOK HAGA OOOO CSO ORO AOAC ICI AO KE
EHO OOKOO OOOOKOROOOGK ddd KOROSOGHHOOGEO OES SGORICA OO III ROKK
JAI EHOO ROO OREO OO OCC GOOK OSS OOOO SORGHHOGO OOO GOO BO ROE ORROOK OOOO IGE KI IOK-+
SOK EHO OOOCC OOG GE OOK BRSSSIOKCE-HOSOORABOO SORE SSS GOK ARO IGRI KKK KOKKOK KE
JOROGSOK-E-FIO BOOK AO OOOO OKO OORICIOKCE-HOORSO OOOOGGSOSA OGGORG OOOO OSORIO IOI KIO
JOKES HORO GOOG GOOO OOGSGES OK BARSG AO OGG SOG RSS GARGS ERGO O GI OK CII
JOO ORO SOCOCHOOO SO OOOO A BGO SOKD ORAS ICROOGO. OGGO REAR ORO GOSS OAK IIE
JO BOO SOOKIGOOKMCHOKE OOECO OOS ONDOO OOKOOGOOO OO GOOG BOOKS OOOG KO OOKS GOROOOO IK KE
JOSKOOO SOOO ORO SSO CHOOBOSO ORO BOGE GOOOOO GOOOKG OOOO KOC OGG AIO CK IORI K
JOO OOD OO SOOROBO OSGHPASOO ROO OOOCHG OOOO KEOOGROG OBE OGE SOKA OBO KOO K IGOR OKO KIO
TTTT
FRET FORO OO RCO OOKOOOO OOOOH HOOUICOOGOC DOOR BGO OOK OS OKS GSO GOGBOROS SSO GIO IO KAKIK IKE
SE OSORIO GEO OHO ORO OOGOOEEHOOREOK HABOOOSO ROBO GCOOGOKCOORICOR GOGO KIO KOSI OIRO IO K IKE
SE OOOO SOS RSIS O OK IOGOSSO AOGOGHO SOKO KORG IAEEO IOGGaE a IOKCECSO GOSS OO IOR IKK KUKOIIOKE
[HOOSIGIOIO OGIO OKGOIOM USO ROR GOCKCEEICHOR ooo GOOGO AROREO OSI EOOK HOGS SIICIOGOIOE IACI IK KKK IOK-E
SEGOOCO OOOO OOK CORO OG SOOO IIGGH HOMO ABH SSAC ORGS CHCRG OARS SKORIK KIKI
SEIKO HOOP OOOO OGGOGG ORGS OOKSOOKGOR IES ASIGOOOKIGOK CORO IGG OOOAOE IG ICO IGIO AION KUKIORIOK OK
SEO GOO ORO S OKGOCOO OK KOOG GOO OOAOO GOGO GCG IHG GaOR GaGOOK FORO OGRA COOK A AO KIK-E
FEO OOCOOOO ORCC OCOD OSIGGGO OKO O GOGO GAO BOGOOOOIGO EOOAO Cao OMCESO ORIG KOKK-E
SPOKE OOOROSOR OOOO BOSE OK TOR GSS OOOO OBGOGOOKGSIBO ORAS SEG ACGIOSISO KOKORO AIOIKIKIK COOKE
FESO SISO OOOO SOOORK BBCO SOO OSSE SSS OKO RCSCO ROOKGG SOA GOOG OCCOBO SIOOAK IOI %K KKK
THOS OOOO ORCC ORO ROCCO OOOO KOSOOOO OOOOOOSG a CSSA aaRACIGORIOOKKLACOSO IOI KOKA
YESS OREO OR OCGO AE ORKIBOO SOK OK ORGS GK ACOSO SSSR aE OBER OSI GRGASIOS OK IOKK KOKIOKAE
HES SO OSS US GOR ORS CE SISOA OEHEC OO RIORESSERA IOORES GOERS ORO. AEE EdF OSGI IOGIa IOI KOKFOKIORE
SHOOK OHSS OSS SOKOKO C ORK FORKS OESS SEO OA ObISS a CEEO KARAS GEaRaCFIAREGO IO OOK OK OOO KOKRACE
SAE EE EEA EEE EEE EEE EEE EEE EEE EEE EEE EE EEE EEE ET TEER EEE EEE EEE EER EEE EE
 potycons 103

d) Plot the sales polygon filled, followed by the expense polygon filled, followed in turn
by the sales polygon again unfilled.
ae ae
+t++4 KK
+tt+4+ JOKOK AK
ttttti ttt FOO OKAK
ttt ttt ttt AOR OOK KK
JO ORIOIIOK RHEE OOK IOICK
EIEIO HOGI IOK EE EEE EEE JO KIOKIGOK
AK
SOOO OO IOI GOK AOK +4 tbttt ttt OOOO KK
FOO OOOO ORK FEEL EFEFE EEE EET EK AK OK KE
AIO IOC OK IOK FERRE EEE EEE EEE EEK KOR OK KK AK
*K SCOR OO OO IOOROIOK be tertteed ee ek F-ai OO AOKAOR,
OOK OK SOOO OOOO IGE + ttt bt eRe EOI OKA
SOO OKKKK SIG OOO IO OO IK IOIK +t ttt EEE EE OK EE EEK OK OK aK
SOK OK GK SAO OO OK FOO OK IKE K+ Et EEE OK EE EE EK OOK
AIO OOK K SOOO OOOO OIOKE + K+ ttt tbat EK EET EO OOK
OOOO AOKI OK SOOO IORI OIOIOIOK IOKOK Eb Eb tt bok bb EK EEE EE EEE OK KOK
SOOO CIOOK KOK JOO OOIOR FORIO KIKI EE EER EEK EEE EE EE KEE EEE EEE EE ET EKO IOK
FOO IO OK ORO IK IKK HOKE EEE EEEE EER EEE EERE EEE EEE EAE E LE ETE OOKK
SOO OR IO OOK OK HII AHHH HIKE FEEL EEF EAL EEE REE EE TELE F EEE EEE SEE tt ek
B++444444 OOOO OKO OKOKGK JOR IO IORI KAO IEEE EE EEE EEE EEE EERE EEE EE EFEE ET EEE TT ETE EK
EEE HOO OOK OOK OK OK RICO IKK IO KK IKK L AEE FEL AEE EEA EE EK EE EEE EEEEE EEF EEE EEE EH Ek
SEER EEE EHO IOI OK HORI KAI KAKI FEF EEEFF FEET EEA EEE EEE FEFE EE EEEE EFT F ETE EE Ek
SEE EHF HII OOOO RK OOK IIIS EE EE EEE EEE EEEE EEE TEETH TET EFT EFE EET TEETH EE PK
SEE EEO EEO IO OOOO OOK IIE EE EEE EPEEE EEE EE EEE EE EEE EE EET EFT EE EE EEE FETE EE EK
SEH HEHE OOOO OKO HOOK IOIOKIOKE TELAT EEE ETE EEEEEEEFEETET AL EEE EEE EET EEE EET EEE EK
FFEFEFERAEE TEE ETE E EIR IK KKK HHH HK EEL EEE EET EE EEEEE ETE EEE EE EEE EPTFE ttt ttt ttt tb tt hk
EEEEE EME TEEPE EEE EE EIR HHH HK AK HIKE TEFL F FLEE ELE ET EF EEE FT ALE TE FETE FETE E TTT P tT tt tt ttt $k
HEEEEEME EE EEE EE EE EEE EE RRA KK KKH LEL EEE ETE EEE EFEA FEET ST EEE EEE EEE EEE EEE EE ET Ft tt tt tk
EEE EME TEER EEE EEE PR EE EK OKI AK KI IK EE EEE EEE EE EEE ET EEE ET EEE FETTEP EE EEE FEET EEE T ETF tt ttt tt $k
HETERFEF FEEL ELE EE EEE EEE EE HK KAM IOKE EEE EEE EEETEETEE EEE EE EEE EEE TEE TELE FEET TET PP tte t ttt bk
EEEKEFEFEF EEE EFF FEET EEE EIKK KI KE EEE E EERE EEE EEE EEE AT ETE EEE TTT EEE E ET ESET ttt ttt ttt ek
FEKETE FEF ETE FEET FETE FAL EF TET ERKKKKKEFE EEE EEE FEET EEF EFET EET E TTT PTT FETE FPP ttt tt ttt ttt ttt tk
TEKEFFEELFETEF EPTFE TFET EEF EFF E RHEE ETE EE TFET EE EFET ETE FEF EFE FETT PTET EE TT EF FFF tt tt ttt ttt ek
EKEFEEEFE EEL E FEET FFE LETTE FEET ELE REEF EEE EEE EEE EET PEEP EEE REET ETE EEE EEE FEF E FET ttt tt ttt tk
HAF FEFE ETE P ETE THE EEEEETEE EEE FETE FEET EPTFE P ETF T PtP ttt ttt ttt ttt tt PEATE AE TEE Et TE PEt tt PEPE EK
MAFFFEFEE FEEL EEE EET EE FEET EEE FEE EEE EEF ETE EEE EE ETE TEA T FEET EP Et ttt TFET EFF d tPF ttt ttt tt tt tt tk
HEFT EELS EEE FETE EEE EE TEEPE EE EE EE EEE EEE TE TET ETT T TE PTET Et Pt Ett ttt ttt tt tht tt ttt ttt tt tt tt hk
HEFEEEEEAE EEE E TEE ELE EEE EFE EE EE EEE TEE EE EEE T ETH T ttt ttt ttt ttt tt tt ttt tt ttt tte ttt tt ttt tt tt hk
MAFEFEFELE EEL E FEET FEES TEEPE ET ETAT EEL EFT ETE EEE TT EE EFEE EEE EF FETT Tt ttt ttt tt ttt ttt tet tt ttt k
AFFEAFEEEEF FEE EL ETE E EES TEFEFELE EEE E EEE A EE EEE ETE TEETH t Et EE tT ET TTT Pt ttt ttt FEET FETE ttt ttt tt x
HELEEEFELEEEFE FELT TFELFET EFF SEE ET ETE T FTE FT Ett ttt ttt ttt tt ttt tt ttt ttt tt tt ttt ttt ttt ttt tt tk
HEFEEFFE LETTE EE EE ATE EFTTA ETHEL ET ETE ETE TET E PET T HEE Ett ttt ttt tt ttt tt ttt ttt ttt ttt ttt ttt ttt tek
HEEEEEEEE EEF EE EEEE EEE E TEE EEE EEE EEE EEE EET EEEE EE EE EE EE EEE EEE TET ETT EEE EEE EEE EEEE EEE EEE EE EEK
HETHEFFEFEEEFEEFE SETHE EEEE EEE EHEEE ESTEE ET ETE EEET EEE TE Ett ttt ttt tt tt ttt ttt ttt ttt ttt ttt
JESSIE OOO COO ORO OOOO OOOOO SOOO OCOG OOOO SIO GOOIOOG CEFR OCOAC OGIO OIE ORI

e) Form a single large polygon from an array of sales figures, followed by expense fig-
ures entered in reverse chronological order and display filled.
ye ++
Kr Kk +K+
1 OOK ++
1 OOK HHO
2 OOOO SHH +
ttttttttt+ 2 OOOO KK» SIO KK +
HII OO KOK+ + 2 DEOOIIOOIOIGIIOK: 210 OOOKOK+
JIC OOK + FOROIOK GOOG OIOOK +
HII OO I+ + JOOS: OOO IO IKE
SHIGEO IOI OO + DIGIOORIOOIOCICS IO FOI IO IK
++ ESO OOOOO OOOO OIG OK OIOOGOIOIOKHE OK IORI IK
xp K++ ZHAICIOOIOIO OOOO ISOS DIOR ISO OOOHOO OIRO ORI
HOKIOK + YEHOIOIOIOIOOOO OOOO R GORI OIC RIIIE IIE
JIGHIOIOIOIOIOK JHICOIIOIOIOO OO OOIOIOIOIGIOK OIG OIOKOOK IORI + ICO IK +
r FEO OK + + ZHICOOI OSG OIORIOIO OOO »— bIOROOKIOKE ICI IO +
{HICK OK JEG OOOO ODIO IOI E TOOK +
HI OIE {HOO III HOOK +4 OOK +
JIC IOI OR IIOK + ICHIC OOOO IOIOIOK 7 $F IOK ++ K+
HIGOOO IORI OI KE JIC IOI OO TOK 9 7 Ho + vot
Toa AGOGO IOC ICE HSIAO IORI ++ ot
OI OIC IORGE + HCCI OOK 9 + 3
TASIORIOOIOIOG ESOC .
OOIKIOIOIOIOIOIG DORIOIODIOIOIOIR OIE DORIC IIOK »
HOOOOOOROOIOOOIOREE HOOK OK yy 5
OOOOOIOOII ;
OGIO OOO RIOIGISIOKIORIOKO FOO IK 97
:
I IO IOI ZOOCIOOO OCI OOOOH OR IOK 9 ¥
OOK OK ZOCOR OO IOI IOK
BICC OO OOOO OK y :
SOOO OK 1K ;
OOK + 2 OOOO OK IOK »
PIII OOO J
DIKE ‘
K+ PICO IO IOK 9K
2 KK 2 DICK » ,
5
r+ OOK 5
rk+ JO OOK»
ot rk
, rok

OP OT TY
POP POPPY POPPOSP OF IPP PUPP POOP PPT TOPO PP EPPO POPOL OP PP OP KOPP OPP OP ES PIO rE POPP OTT HIPHOP
104 CHAPTER THREE

3-9 a) Write a subroutine to drawa filled square with side length 0.1 using only relative com-
mands. Then, by positioning the pen and calling this subroutine, construct a checkerboard.
b) Construct other patterns by repetition of different polygan shapes.
3-10 a) Use rectangular polygons to form a bar graph of data in Programming Problem 3-8.
b) Display both widget sales and expense data on the same bar graph by using bars with
different interior and edge styles for the two types of data. Place both bars for each month on top
of each other, but enter the larger of the two first.

a eee
Aas eee
ae ees
se ees
Cabs eta eG Le moras aS wee
wee eee St el RM eee
eae wee = lea eee oe
woe eee om ts cee aes
eee ame poate ar aes ues
ANS eon wee r — mee eee ses
eee see aoe toee ——e rete see eee
see eee eg! tenes e = ens ese
wae e+e wes! tees 5 ees see eee
eee eee wee ‘eesr oe S54 ser oes
weet ores eee ‘oes we == one
aan ae ous one eis a oon fo aes
wee uae nee ‘eas 4 <1 ene = oes
oes one eee eee see =~ ese —— ose
exe wen ane eae =a = ae Ss oes
xem «ee eee " = wee SEAve ois
= = ake ee wee! — = eee ees wae =
= ase eee ox ee! a 255 ses eee ee ==
ee = nee ee wee! ee aoe es =
6 x oe wee eee! = FS eee see = -s-
a ees ven = ~ ese eee —— am
= VERN VSS ON SRN a Vat ASN el Gee C Om ee Helos Ue
os wee xan een et Ext ae wen wee erate eered
a ace eee eas ot Le ae eee wee wa ar —
= wee . eee od ap Veh = eee eee eases ase
a = ae es uae acai} =e) Je ‘one wee aan aad
freee eee aoe one ee, oe, —— = aoe een cate =-=
es eee ae as eee SS as et! ee eae = ae
= * ox <—= uae anor ete wee == en
Se SS Set RR it Sl eT Nema Aimee) Ue oe Woon
eee —— aa ae as wee eee =~ =
em nae = — ee == 3 = wee ees ow =e
ae een so ee we oe Veto wee sae irate oe
** «* = = = ss 5 tear tees eee —
ae eee or a eee Nee see wee caret Set
«ee eee a & = fare? ane wee wee ee =
eK ek ae fa ee Sat rare nee eae eS ae
aut woe = = 63 SoS mma A er toe ene a
uae eae == en ee Fess a ea eas eee cafe a
WAM NAT VW GT ANE MNGIRE Ye WAN Sa SRLNR DM ELC Poe CaeNRceClee BIGADDN NTT Sun RTAL Belt

3-11 Write a graphing procedure which takes an array of values and plots them in a bar graph.
3-12 Write a routine WEDGE(A) which draws a polygon in the shape of a wedge of pie. The
wedge should have radius 1 and one edge on the x axis with the point at.the origin. The angle of
the wedge is A radians. A round border is approximated with line segments connecting points
on
a circle. The ith border point is given by Xi = COS(Ai), Yi = SIN(Ai), where Ai = (i * A)/N
and N = INT(4 « A). The index i will step between O and N, inclusive.

(0, 0) (1,0)
 potycons 105

3-13 Write a routine called THICK-LINE-ABS-2(X, Y, W) which draws a thick

line of width W
centered on the line segment between the current position and the point (X, Y). The
thick line
will actually be a polygon. Design the polygon so that the ends of the thick line are
perpendicu-
lar to its direction.

w (DF — PEN
— X, DF — PEN
— Y)

3-14 Extend your graphics system to allow the user to specify whether or not polygon borders
should be drawn.
3-15 We allocated opcodes for 16 line styles and 16 fill styles. This may not be enough. A
graphics system may have 64 gray levels or 256 colors plus thicknesses and patterns. We can ac-
commodate this variety by assigning generic opcodes (line style, line color, fill style, fill color,
etc.) and using the operand field (DF-X or DF-Y) to hold the particular style value. Modify your
graphics system to do this.
*3-16 When using a frame buffer to display polygons, the most recently drawn polygon can
obscure anything drawn prior to it. We can use this to give a feeling of three-dimensional depth
by drawing the background objects prior to the foreground. Develop routines which allow the
user to enter each polygon with some depth or priority number. These polygons should then be
sorted into priority order so that the most distant (lowest priority) polygons are drawn before the
closest (highest priority).
*3.17 Construct a font of large characters and/or digits out of filled polygons. Write a separate
procedure for each symbol, and use only relative commands so that the position of the symbol
may be changed.
106 CHAPTER THREE

*3-18 Extend the polygon-filling algorithm to provide antialiasing. Do this by in effect doubling
the resolution in both x and y directions. Set up a storage buffer for two double-length scan
lines. As you step through the polygon-filling algorithm, fill inthese buffers with the back-
ground values from the display (replicating values to match the higher resolution). Next, fill the
pair of scan lines with the polygon. Finally, average the intensities of the four-pixel high-resolu-
tion cells to arrive at an intensity value for the corresponding pixels on the actual display.

*3-19 Some graphics hardware will directly support the drawing of trapezoids with their parallel
sides in the scan-line direction. Write a program which will decompose arbitrary polygons into
such trapezoids.
*3-20 We often find convex polygons easier to deal with than concave ones. Triangles are
par-
ticularly convenient because they are guaranteed to be both convex and planar. Write a program
which will decompose arbitrary polygons into triangles.
*3-21 If instead of filling in a polygon, our scan-line program just set the edge pixels,
we would
have a program which constructs a line drawing from top to bottom. This is
just the order that
some raster displays (and line printers) need the pixel information in for display.
Now imagine
that the instructions not only for a single polygon are sorted, but also those
for the entire display
file. Then pixel settings for the entire picture can be generated in just the order
in which the dis-
play device needs them. By correctly sorting the display-file instructions,
it is possible to avoid
the need for the large frame buffer. Assume that the display contains
only MOVE and LINE
commands. Write DOLINE, DOMOVE, and DISPLAY routines, modeled
after FILL-POLY-
GON and its friends, which eliminate the need for a frame buffer array
for line printer output.
 CHAPTER

TRANSFORMATIONS

INTRODUCTION
One of the major assets of computer graphics is the ease with which certain alterations
of the picture may be performed. The manager can alter the scale of the graphs in a re-
port. The architect can view a building from a different angle. The cartographer can
change the size of a chart. The animator can change the position of a character. These
changes are easy to perform because the graphic image has been coded as numbers and
stored within the computer. The numbers may be modified by mathematical operations
called transformations.
Transformations allow us to uniformly alter the entire picture. It is in fact often
easier to change the entire computer-drawn image than it is to alter only part of it. This
can provide a useful complement to hand drawing techniques, where it is usually easier
to change a small portion of a drawing than it is to create an entirely new picture.
In this chapter we shall consider the geometric transformations of scaling, trans-
lation, and rotation. We shall see how they can be simply expressed in terms of matrix
multiplications. We shall introduce homogeneous coordinates in order to uniformly
treat translations and in anticipation of three-dimensional perspective transformations.
The algorithms presented in this chapter will describe two-dimensional scale, transla-
tion, and rotation routines.

MATRICES
Our computer graphics images are generated from a series of line segments which are
represented by the coordinates of their endpoints. Certain changes in an image can be
easily made by performing mathematical operations on these coordinates. Before we

107
108 CHAPTER FOUR

consider some of the possible transformations, let us review some of the mathematical
tools we shall need, namely, matrix multiplication.
For our purposes we will consider a matrix to be a two-dimensional array of
numbers. For example, —

10 14 93 1 10.2
01 4 5 6 vl0 Ode
OmrOre!

are four different matrices.

Suppose we define the matrix A to be
23
Avo tae 5: 26 (4.1)
Tai ao

Then the element in the second row and third column would be A(2, 3) and would
have the value 6.
The matrix operation which concerns us most is that of multiplication. Matrix
multiplication is more complex than the simple product of two numbers; it involves
simple products and sums of the matrix elements. Not every pair of matrices can be
multiplied. We can multiply two matrices A and B together if the number of columns
of the first matrix (A) is the same as the number of rows of the second matrix (B). For
example, if we chose the first matrix to be the matrix A defined in Equation 4.1 and
matrix B to be

boy
Boo) ) (4.2)
0 =N
©

then we can multiply A times B because A has 3 columns and B has 3 rows. Unlike
multiplication of numbers, the multiplication of matrices is not commutative; that is,
while we can multiply A times B, we cannot multiply B times A, because B has only 2
columns (which does not match the 3 rows of A). When we multiply two matrices, we
get a matrix as a result. This product matrix will have the same number of rows as the
first matrix of the two being multiplied and the same number of columns as the second
matrix. Multiplying the 3 x 3 matrix A times the 3 X 2 matrix B gives a3 X 2 matrix
result C. Ra
The elements of the product matrix C are given in terms of the elements of ma-
trices A and B by the following formula:

C(i, k) = 3 AG,
j)BG, k) (4.3)
For our particular example of C = AB

Cs 1
(4.4)
conn
Nhe OAW eNO
ore
 TRANSFORMATIONS 109

the element C(1, 1) is found by multiplying each element of the first row of A by the
corresponding element of the first column of B and adding these products together.

Cd, 1) = AC, DBC, 1) a A(1, 2)B(2, 1) + AC, 3)B(3, 1)

= (OU zt) Soy = = 1 (4.5)
The element C(3, 2) would be
C(3, 2) = A(3, 1I)BC1, 2) + AQ, 2)B(2, 2) + AQ, 3)BQ3, 2)

= (7)) + (8)(2) + (9)(1) = 25 (4.6)

Performing this arithmetic for every element of C shows us that
=F. 7
Ce tH 1 16 (4.7)
= 14s
Multiplication is associative. This means that if we have several matrices to multiply
together, it does not matter which we multiply first. In mathematical notation:
A(BC) = (AB)C (4.8)
This is a very useful property; it will allow us to combine several graphics transforma-
tions into a single transformation, thereby making our calculations more efficient.
There is a set of matrices with the property that when they multiply another ma-
trix, they reproduce that matrix. For this reason, the matrices in this set are called iden-
tity matrices. They are square matrices (same number of rows and columns) with all the
elements 0 except the elements of the main diagonal, which are all 1. For example,
IeaiOx -0
|1 | ‘ : 0 10
001

and so on.
We can see that if

a L020
b= 16~-1.- 6 (4.9)
ORLOr 1

then
A= AI (4.10)

SCALING TRANSFORMATIONS
Now how does all this apply to graphics? Well, we can consider a point P; = [x; yj]
as being a 1 X 2 matrix. If we multiply it by some 2 X 2 matrix T, we will obtain
another 1 X 2 matrix which we can interpret as another point
[xX. y2] = Pp = P,T (4.11)
110 = cHaPTER FOUR

Thus, the matrix T gives a mapping between an original point P, and a new point
P,. Remember that our image is stored as a list of endpoints. What will happen if we
transform every point by means of multiplication by T and display the result? What will
this new image look like? The answer, of course, depends upon what elements are in
matrix T. If, for example, matrix T were the identity matrix

Te A A (4,12)
then the image would be unchanged.
If, however, we choose T to be T,

T-|3 a (4.13)
then,

[x2 fe 1
yo2J=[x yr] 2 OTe=[2x1
01 yi] (4.14)

Every new x coordinate would be twice as large as the old value. Horizontal lines
would become twice as long on the new image. The new image would have the same
height, but would appear to be stretched to twice the width of the original. (See Figure
4-1.)
The transformation matrix

hae 0.5 0
Orsaik (4.15)
would shrink all x coordinates to one-half their original value. The image would have
the original height but would be squeezed to one-half the width.

Before

FIGURE 4-1
After Scaling x coordinates by 2.
 TRANSFORMATIONS 111

Now, if we were to stretch the image to twice the width and then compress it to
one-half the new width,

P, = (P\T,)T, (4.16)
we would expect to get the original image back again. Let us check that this is so by
multiplying the two transformations T, and T, together first. The associative property
for matrix multiplication allows us to write

P. = P,(T,T2) (4.17)
Multiplying the two transformations together combines them into a single transforma-
tion .

TiT2 — (4.18)
1
0) 05 2 1

But this is just the expected identity matrix, which will not change an image.
We can make an image twice as tall with the same width by finding a transforma-
tion which just multiplies the y coordinate by 2. Such a transformation is given by
1 0
Tie ee (4.19)
Multiplying an arbitrary point by this matrix shows that this is true. (See Figure 4-2.)

[x2 yo] =[% yr] ‘ >|= [x; 2yi1] (4.20)

By applying both transformations T, and T3, we will make the image both twice
as wide and twice as tall. In other words, we would have a similar image, only twice
as big.

a
r

Before

FIGURE 4-2
After Scaling y coordinates by 2.
112 ~~ cHapTER FOUR

P, = P,T,T3 (4.21)

Again, by multiplying the two transformation matrices together, we can obtain a single
transformation matrix for making the entire image twice as large. (See Figure 4-3.)

Torre ten
20 -Awid o|- 2; 4 (4.22)
In general, transformations of the form
Ss. 0
S= (4.23)
0 Sy

change the size and the proportion of the image. They are called scaling transformations.
S, is the scale factor for the x coordinate and s, for the y coordinate.
Note that when we scale the image, every point except the origin changes. This
means that not only the size of an image will change but also its position. A scale in x
by a factor greater than 1 will cause the image to shift to the right, along with making
it wider. A scale in x by a factor less than 1 will shift the image to the left. A scale in y
will shift the image up and down as well as change its height.

SIN AND COS

The next transformation we would like to consider is that for rotation. To prepare for
the discussion of rotations we shall review some basic trigonometry. Suppose we have
a point P; = (x,, y,) and we rotate it about the origin by an angle @ to get a new posi-
tion P, = (X2, yz). We wish to find a transformation which will change (x,, y) into
(Xj, Y2). But, before we can check any transformation to see if it is correct, we must
first know what (x2, y2) should be in terms of (x, y,) and @. To determine this we shall
need the trigonometric functions sine and cosine (abbreviated sin and cos). We can de-

Before

FIGURE 4-3
After Scaling both x and y coordinates.
 TRANSFORMATIONS 113

fine sin and cos for an angle @ in the following manner. Let us draw a line segment
from the origin at the angle @ counterclockwise from the x axis, and suppose that the
line segment we have drawn has length L. (See Figure 4-4.)
The line segment will then have endpoints (0, 0) and (x, y) and length
L= (x? a a dag

Then, the ratio of the height of the (x, y) endpoint above the x axis (the y-coordinate
value) and the length of the segment will be the sine of the angle

: y
sin.o.— fea (4.24)

and the ratio of the distance to the right of the y axis (the x-coordinate value) and the
length of the segment will be the cosine of the angle
X
cos 8 = Ge+yy (4.25)

Note that if we draw a segment with length L = 1, then

sin§ = y and cos 6 = x (4.26)

ROTATION
To determine the form for the rotation transformation matrix, consider the point (1, 0). If
we rotate this point counterclockwise by an angle @, it becomes (cos 9, sin 8) (see Fig-
ure 4-5), so

[cos@ sing]= ja 0| : a= [a (4.27)

If we rotate the point (0, 1) counterclockwise by an angle 6, it becomes (-sin 8, cos 8).
(See Figure 4-6.)

, [-sin 6 cos 0] = |0 1| . al |ea| (4.28)

y axis
(x, y)

3 FIGURE 4-4
X axis Sata
(0, 0) Definition of angle.
114. CHAPTER FOUR

sin 8

an FIGURE 4-5
3) cos8 Rotating the point (1, 0).
From these equations we can see the values of a, b, c, and d needed to form the rota-
tion matrix. The transformation matrix for a counterclockwise rotation of 0 about
the
origin is

cos@ sin@
Re -sing@ cos9 (4.29)

FIGURE 4-6
—sin 0 (0,0)
Rotating the point (0, 1).
 TRANSFORMATIONS 115

As an example, suppose we wished to rotate the point (2, 3) counterclockwise by an

angle of 1/6 radians. (See Figure 4-7.) Then the rotation matrix would be

cos sin = | se ths

|ae0.866 0.5 | (4.30)
-sin® cos -0.5 0.866|
and the rotated point would be

|2 3| SR0.866 0.5 |°.32

-/02 3.598 | (4.31)
We can rotate an entire line segment by rotating both the endpoints which specify it.
The sign of an angle determines the direction of rotation. We have defined the ro-
tation matrix so that a positive angle will rotate the image in a counterclockwise direc-
tion with respect to the axes. In order to rotate in the clockwise direction we use a
negative angle, so the rotation matrix for an angle 8 clockwise would be

Ro |Mine) eoe-8) oe
or since
cos (-9) = cos 0 (4.33)

and ~
sin (-@) = —sin 0 (4.34)
this may be rewritten as

Roba ane| a9
for clockwise rotations about the origin.

FIGURE 4-7
Rotation of (2,3) by 7/6.
116 CHAPTER FOUR

HOMOGENEOUS COORDINATES
AND TRANSLATION
Suppose we wished to rotate about some point other thansthe origin. If we had some
way of moving the entire image around on the screen, we could accomplish such a ro-
tation by first moving the image until the center of rotation was at the origin, then per-
forming the rotation as we just discussed, and finally, moving the image back where it
belongs.
Moving the image is called translation. It is easily accomplished by adding to
each point the amount by which we want the picture shifted. If we wish the image
shifted 2 units to the right, we would add 2 to the x coordinate of every point. To move
it down 1 unit, add —1 to every y coordinate. (See Figure 4-8.)
In general, in order to translate the image to the right and up by (t,, t,), every
point (xj, y;) is replaced by a new point (x5, y2) where
Xou— sXe EG Ya. Sy Sant dy (4.36)

Unfortunately, this way of describing translation does not use a matrix, so it cannot be
combined with other transformations by simple matrix multiplication. Such a combina-
tion would be desirable; for example, we have seen that rotating about a point other
than the origin can be done byatranslation, a rotation, and another translation. We
would like to be able to combine these three transformations into a single transforma-
tion for the sake of efficiency and elegance. One way of doing this is to use homoge-
neous coordinates. In homogeneous coordinates we use 3 X 3 matrices instead of 2 X 2,
introducing an additional dummy coordinate w; points are specified by three numbers
instead of two. The first homogeneous coordinate will be the product of x and w, the
second will be the product of y and w, and the third will just be w. A coordinate point
(x, y) will be represented by the triple (xw, yw, w). The x and y coordinates can easily
be recovered by dividing the first and second numbers by the third. We will not really
use the extra number w until we consider three-dimensional perspective transforma-

Before

FIGURE 4-8
After Translation.
 TRANSFORMATIONS 117

tions. In two dimensions its value is usually kept at | for simplicity. Still, we will dis-
cuss it in its generality in anticipation of the three-dimensional transformations.
In homogeneous coordinates our scaling matrix
Ss, 0
0 Sy

becomes
Ss 0 0
S=| 0 sy 0 (4.37)
O; Orn)
If we apply this to the point (xw, yw, w), we obtain
s 0 0
xw yw w O Sy 0} =| 8xxw Syyw w (4.38)
er, 0 0 1
fete OKA, free
Dividing by the third number w gives |

(S,X, Syy)
which is the correctly scaled point.
The counterclockwise rotation matrix
- cos@ sin @
-sin®@ cos@
becomes, using homogeneous coordinates,

cosq@ sing 0
R=]-sing cos@ 0 (4.39)
0 0 1

Applying it to the point (x, y) with homogeneous coordinate (xw, yw, W) gives
cosg sind 0
[xw ‘yw w]|-sin@ cos6 O);=
0 0 1

[(xw cos @— yw sin®) (xw sin@+ywcos®@) w] (4.40)

for the correctly rotated point

(x cos 6-y sin 6, x sin@ + y cos8)
ty is
The homogeneous coordinate transformation matrix for a translation of t,,
| ee a
Pag 2 oO (4.41)
tx ty 1

To show that this is so, we apply the matrix

118 CHAPTER FOUR

eo
[Rm%,, YW dich Dinesh = [(xw+txw) (yw+tyw) wl] (4.42)
tx t “< OO
-
.

for the translated point.(x + t,, y + ty)

COORDINATE TRANSFORMATIONS
We have shown how applying a transformation to a point yields a new point, but trans-
formations may also be used to change coordinate systems. For example, a distance
measured in inches can be converted to the same distance measured in centimeters by
means of a scale. The actual operation of the transformation is the same as already de-
scribed; only the interpretation changes. The coordinates of the point, when multiplied
by the transformation, represent the same point, only measured in different coordi-
nates.
Translations are useful coordinate transformations when the origins are not
aligned. For example, we think of the lower-left corner of the display as being the ori-
gin (0, 0), but in some display systems this point may actually correspond to pixel
(1, 1). Another example is a display which places the (0, 0) pixel in the upper-left
corner and numbers the scan lines from top to bottom. This is sometimes done on al-
phanumeric printers because it is the order in which the lines are printed, and on raster
displays because it is the order that they are actually scanned. To convert between these
coordinates and the ones we have been using, we need ascale of | in x but -1 in y to
reverse the scan-line order, and also a translation in y by the vertical screen dimension
to move the origin to the proper corner.
Rotations may also be used in coordinate transformations, but are usually for
angles of 1/2 (90 degrees). For example, a printer using 8.5 by 11 inch paper may
have the y axis along the long edge and the x axis along the short edge. This is called
portrait mode. We might prefer to orient the y axis along the short edge and the x axis
along the long edge (perhaps so that the orientation of the paper is a better fit to the
shape ofa television’s display). This is termed landscape mode. A rotation of 7/2 and a
translation to reposition the origin in the lower-left corner will do this.
We have already seen an example of a coordinate transformation, although we
did not use transformation-matrix notation. The example is the transformation from
normalized device coordinates to actual device coordinates. The arithmetic we used to
do the conversion was

X1 <— X * WIDTH + WIDTH-START;

Y1 <—Y * HEIGHT + HEIGHT-START;

We can see now that this is a scale by WIDTH for x and HEIGHT for y, followed by a
translation by WIDTH-START and HEIGHT-START. The full transformation matrix
for this change in coordinates is

WIDTH 0 0
D= 0 HEIGHT 0 (4.43)
WIDTH- START HEIGHT-START 1
 TRANSFORMATIONS 119

ROTATION ABOUT AN ARBITRARY POINT

Now let’s determine the transformation matrix for a counterclockwise rotation about
point (Xc, Yc). (See Figure 4-9.)
We shall do this by three transformation steps. We shall translate the point (xc,
Yc) to the origin, rotate about the origin, and then translate the center of rotation back
where it belongs. (See Figure 4-10.)
Matrix multiplication is not commutative. Multiplying A times B will not always
yield the same result as multiplying B times A. We must be careful to order the ma-
trices so that they correspond to the order of the transformations on the image. We shall
place the coordinates of the point on the left and the transformation matrix on the right.
With this ordering, if an additional matrix product is introduced on the right (post-mul-
tiplication), then the corresponding transformation will be carried out after the original
transformation. If an additional matrix product is introduced on the left of the original
transformation matrix (between it and the coordinates of the point), then the new trans-
formation takes place prior to the original transformation. Multiplying on the left is
called pre-multiplication. We use post-multiplication in the construction of our general
rotation.
The translation which moves (Xc, yc) to the origin is
1 O.'0
T21+°0%.-1 Y0 (4.44)
rheitn veut
the rotation is

cos@ sing 0
R=|-sing cos@ 0 (4.45)
nat

and the translation to move the center point back to its correct position is
ee 0.10
Pee Ot 0 (4.46)
Xe aye

FIGURE 4-9
Rotation about an arbitrary point.
120 = cHAPTER FOUR

Translate Rotate Translate back

FIGURE 4-10
Three steps in the rotation about an arbitrary point.

To transform a point, we would multiply

(([xw yw w]T,)R)T>)
but we reassociate and multiply all the transformation matrices together first to form an
overall transformation matrix

[xw yw w)(T\(RT>))

1 0.10 cosg@ sing 0 er Or wa

TeR: Ty=|2-00 1 Oj} |-sing cos@ 0 Cis ts 28
“Xo -¥o 1 0 O shititealer a
1 0 O cosg@ sing 0O
=| 0 1 O|]-sing cose 0
=X%_¢ Yel Xe- Yo. 1

cos 6 sin @ 0
ahelg Sie
-X¢cos@ + YcSMQ +X¢
niceSOS8
-XcSiIN@-Yecos@+Yc
01 (4.47)
This is the overall transformation for a rotation by @ counterclockwise about the point
(Xc, Yc). Note that this matrix may also be formed by an initial rotation of @, followed
by a single translation by the values in its third row. “2

OTHER TRANSFORMATIONS
The three transformations of scaling, rotating, and translating are the most useful and
the most common. Other transformations are also possible. Since any 2 X 2 transfor-
mation matrix
a b
Card
can be converted to a3 X 3 homogeneous coordinate matrix as

fp
onlone
-OO
 TRANSFORMATIONS 12]

we will present only the 2 x 2 form for some of these transformations:

=e 0
rk | reflection in the y axis (see Figure 4-11)

1 0O
0 -1 reflection in the x axis (see Figure 4-12)

-1 O
0 -1 reflection in the origin (see Figure 4-13)

|o 1
1 0 reflection in the line y = x (see Figure 4-14)

0 -1
-1 O reflection in the line y = —x (see Figure 4-15)

la
01 y shear (see Figure 4-16)

1.0
b 1 x shear (see Figure 4-17)

The first three reflections are just scales with negative scale factors. The reflections in
the lines y = x and y = -x can be done byascale followed bya rotation.
The shear transformations cause the image to slant. The y shear preserves all the
x-coordinate values but shifts the y value. The amount of change in the y value de-
pends upon the x position. This causes horizontal lines to transform into lines which
slope up or down. The x shear maintains the y coordinates, but changes the x values
which causes vertical lines to tilt right or left. It is possible to form the shear transfor-
mations out of sequences of rotations and scales, although it is much easier to just
form the matrix directly. It is also possible to build rotation and some scaling transfor-
mations out of shear transformations.

Before After

FIGURE 4-11
Reflection in the y axis.
122 CHAPTER FOUR

Before After

FIGURE 4-12
Reflection in the x axis.

INVERSE TRANSFORMATIONS
We have seen how to use transformations to map each point (x, y) into a new point (x’,
y’). Sometimes, however, we are faced with the problem of undoing the effect of a
transformation; given the transformed point (x’, y’), we must find the original point
(x, y). An example occurs when the user indicates a particular position on a displayed
image. The display may show an object which has undergone a transformation. If we
want to know the corresponding point on the original object, then we must undo the
transformation.
Undoing a transformation appears to be a transformation itself. Given any point
(x, y’), we need a way of calculating a new point (x, y). Is there perhaps a transforma-
tion matrix which will do this? Often there is, and it can be determined by matrix inver-

Before After
FIGURE 4-13
Reflection in the origin.
 TRANSFORMATIONS 123

Before After
FIGURE 4-14
Reflection in the line y = x.

sion. The inverse of a matrix is another matrix such that when the two are multiplied to-
gether, the identity matrix results.
If the inverse of matrix Tis T”!, then

t= Too (4.48)
To see that this is what we need, consider Equation 4.11 which transforms point P, to
yield point P,. If we multiply both sides of this equation by the inverse of transforma-
tion matrix T, we get
Dis Pan: =P = P, (4.49)
This shows that the inverse of T transforms P, back into P).
This inverse of a matrix can be a handy thing to have, so we shall show how to
find it. We shall do so in terms of another matrix calculation called the determinant. The

Before After

FIGURE 4-15
Reflection in the line y = —x.
124 = CHAPTER FOUR

C) | }

FIGURE 4-16
Before After y shear.

determinant of a matrix is a single number calculated from the elements of the matrix.
For a single element matrix, the determinant is simply the value of the element.

det|t] = t (4.50)

For larger matrices, we express the determinant as a combination of the determinants

of smaller matrices. The smaller matrices are called the minors and are made by remov-
ing a row and column from the larger matrix. Call Mj; the matrix formed by removing
row i and column j from matrix T. Then the determinant of T is given by

det T; = 2 t,(—1)' +) det M, (4.51)

This says that we pick some row of the original matrix T and multiply each element in
the row by the determinant of the minor for that position, alternating signs. The sum of
these numbers is the full determinant.
The determinant of a 2 x 2 mattix is

ti1 ti
det to: t22 = tyto2 — ty te; (4.52)

The determinant of a 3 X 3 matrix is

det T = ty;(tyat33 — ty3tz2) —tya(tot33 — to3tz,) + ty3(toit32 —toots,) (4.53)

The determinant of the homogeneous coordinate transformation matrices we usually
deal with is :

FIGURE 4-17
Before After x shear.
 TRANSFORMATIONS 125

det = ae — bd (4.54)
oop roo
—->OoaQ

Now we can express the inverse of a matrix in terms of determinants.

si cpl durc et Me
U;. 4.55
: det T ( )

where t;; is an element of the inverse of matrix T. Note that the order of i and j is re-
versed on the minor.
The inverse of the homogeneous coordinate transformation matrix is

asida.0 | e -d 0
inv|b
e O|]= bd -b a 0 (4.56)
Crt c= bf-ce cd-af ae-bd

TRANSFORMATION ROUTINES
Now that we have seen the mathematics behind transformations, let’s look at some al-
gorithms to actually perform them. We shall construct routines for translating, rotating,
and scaling. The routines that create the transformations will modify a homogeneous
coordinate transformation matrix. This matrix can then be applied to any point to ob-
tain the corresponding transformed point. While we use the notion of homogeneous
coordinates, we shall not actually store the third coordinate w; instead, we shall ar-
range our transformations so that w will always be 1. We can therefore just use |
whenever we need the coordinate w. For the same reason, we shall not store the last
column of the transformation matrix. This column would always contain 0, 0, 1; and
since we know this, we can avoid actually storing these numbers. We shall save the
3 X 3 homogeneous transformation matrix in the 3 x 2 array named H.
We begin with a routine to set the transformation matrix to the identity matrix.
This clears the transformation so that we can start fresh.

4.1 Algorithm IDENTITY-MATRIX (H) Routine to create the identity transformation

Argument _His a transformation array of 3 x 2 elements.
Local I, J variables for stepping through the H array
BEGIN
FOR I = 1TO3 DO
FOR J = 1TO2 DO
IFI = J THEN H[{I, J] <1
ELSE H{I, J] —0;
RETURN;
END;

The next algorithm causes ascaling transformation. It has the effect of multiply-
ing the matrix H on the right by a scaling transformation matrix of
126 CHAPTER FOUR

SX, G08 0
OFS ¥e0
OF T0381
a)

We notice that there are a lot of zeros in this matrix, so if we follow the multiplication
steps of Equation 4.3, we will be multiplying and adding many noncontributing terms.
The algorithm given here avoids this wasted effort by only dealing with the nonzero
terms.

4.2 Algorithm MULTIPLY-IN-SCALE (SX, SY, H) Routine to post-multiply the trans-

formation matrix by a scale transformation
Arguments SX is the x scale factor
SY is the y scale factor
H is a3 X 2 transformation matrix
Local I for stepping through the array
BEGIN
FOR I = 1TO3 DO
BEGIN
H[I, 1] —H[I, 1] * SX;
H[I, 2] —H[I, 2] « SY;
END;
RETURN;
END;

For translation TX, TY, we post-multiply H by the translation matrix

Fue Oe 4)
Oy aa
FA WT Yo "1
Again we simplify the multiplication by neglecting zero terms. Since the third column
of the transformation matrix is always 0, 0, 1, the algorithm for translation is as fol-
lows.

4.3 Algorithm MULTIPLY-IN-TRANSLATION (TX, TY, H) Routine to post-multi-

ply the transformation matrix by a translation
Arguments TX translation in the x direction
TY translation in the y direction
H a3 X 2 transformation matrix
BEGIN
His, 1) << His xe
B32) HS 2 bye
RETURN;
END;

For a rotation of A radians counterclockwise, we post-multiply by

cosA sinA 0
-snA cosA 0
0 0 1
 TRANSFORMATIONS 127

The algorithm takes the angle as an argument, calculates the sine and cosine, and then
performs the matrix multiplication for nonzero terms.

4.4 Algorithm MULTIPLY-IN-ROTATION(A, H) Routine to post-multiply the trans-

formation matrix bya rotation
Arguments A angle of counterclockwise rotation
H a3 X 2 transformation matrix
Local S, C the sine and cosine values
I for stepping through the array
TEMP temporary storage of the first column
BEGIN
C <— COS(A);
S — SIN(A);
FOR I = 1TO3 DO
BEGIN
TEMP < H{[I, 1] * C—H[I, 2] « S;
H[I, 2] — Hi 1) * S + HUI, 2] + C;
H[I, 1] < TEMP;
END;
RETURN;
END;

The above routines serve to create a transformation matrix, but we still need a
routine to apply the resulting transformation. The following routine transforms asingle
point. The point coordinates are passed to the subroutine as arguments, and the trans-
formed point is returned in the same variables.

4.5 Algorithm DO-TRANSFORMATION(X, Y, H) Routine to transform a point

Arguments X, Y the coordinates of the point to be transformed
H a3 X 2 transformation matrix
Local TEMP temporary storage for the new X value.
BEGIN
TEMP <—X « H[1, 1] + Y* H[2, 1] + H[3, 1);
Y —X « H[l, 2] + Y * H[2, 2] + H{3, 2);
X < TEMP;
RETURN;
END;

To perform a transformation, we must form the appropriate transformation ma-

trix and then apply it to the points in our display. There are several ways this might be
done. In our approach we shall deny the user access to the MULTIPLY-IN-SCALE,
MULTIPLY-IN-TRANSLATION, and MULTIPLY-IN-ROTATION routines. The user
will therefore not be able to build up complex transformations by multiplying several
scales, translations, or rotations (if he wishes to do this, he will have to write his own
routines). Instead, the user will be allowed only one scale, one rotation, and one trans-
lation, to be applied in that order. The transformations will be applied as the display
file is interpreted. The user can change the values of these transformations at any time,
but the image will be formed using the values in effect at the time the display file is in-
128 cHAPTER FOUR

FIGURE 4-18
The image transformation.

terpreted. The following three algorithms describe user routines for saving the transfor-
mation parameters until it is time to interpret the display file.

4.6 Algorithm TRANSLATE(TX, TY) User routine to set the translation parameters
Arguments TX, TY the translation amount
Global TRNX, TRNY storage for the translation parameters
BEGIN
TRNX < TX;
TRNY <TY;
CALL NEWFRAME;
RETURN;
END;

4.7 Algorithm SCALE(SX, SY) User routine to set the scaling parameters
Arguments SX, SY the scaling factors
Global SCLX, SCLY storage for the scale parameters
BEGIN
SCLX <— SX;
SCLY — SY;
CALL NEWFRAME;
RETURN;
END;

4.8 Algorithm ROTATE(A) User routine to set the rotation angle

Argument Athe rotation angle
Global ANGLa place to save the rotation angle
BEGIN
ANGL <A;
CALL NEWFRAME;
RETURN;
END;

We now need a routine which will take the transformation parameters specified
by the user and use them to form a transformation matrix. This is just a matter of pass-
ing the values to the transformation matrix multiplication routines which we defined
above. (See Figure 4-18.)

4.9 Algorithm BUILD-TRANSFORMATION Routine to build the image transforma-

tion matrix
Global ANGL, SCLX, SCLY, TRNX, TRNY the transformation parameters
IMAGE-XFORM a 3 X 2 array containing the image transformation
 TRANSFORMATIONS 129

BEGIN
IDENTITY-MATRIX(IMAGE-XFORM);
MULTIPLY-IN-SCALE(SCLX, SCLY, IMAGE-XFORM);
MULTIPLY-IN-ROTATION(ANGL, IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(TRNX, TRNY, IMAGE-XFORM);
RETURN;
END;

To apply these transformations to a picture, we shall modify three of the: al-

gorithms from previous chapters. We shall extend the MAKE-PICTURE-CURRENT
algorithm to include a call on BUILD-TRANSFORMATION. Thus the parameter set-
tings at the time the MAKE-PICTURE-CURRENT routine is executed will be the
values used for the transformation.

4.10 Algorithm MAKE-PICTURE-CURRENT (Algorithm 2.12 revisited) User rou-

tine to show the current display file
Global FREE the index of the next free display-file cell
ERASE-FLAG indicates if frames should be cleared
BEGIN
IF ERASE-FLAG THEN
BEGIN
ERASE;
ERASE-FLAG < FALSE;
END;
BUILD-TRANSFORMATION;
IF FREE > 1 THEN INTERPRET(1, FREE — 1);
DISPLAY;
FREE < 1;
RETURN;
END;

Each point retrieved from the display should be multiplied by the transformation
matrix, and the resulting product should be used. This can be accomplished by modify-
ing the INTERPRET and LOAD-POLYGON routines to call GET-TRANSFORMED-
POINT instead of GET-POINT. (See Figure 4-19.)

4.11 Algorithm INTERPRET(START, COUNT) (Algorithm 3.6 revisited) Scan the

display file performing the instructions
Arguments START the starting index of the display-file scan
COUNT the number of instructions to be interpreted
Local NTH the display-file index
OP, X, Y the display file instruction
BEGIN
a loop to do all desired instructions
FOR NTH = START TO START + COUNT — 1 DO
BEGIN
GET-TRANSFORMED-POINT(NTH, OP, X, Y);
IF OP < -31 THEN DOCHAR(OFP, x, Y)
130 CHAPTER FOUR

Image
transform
data
User program
Display
file

Figure 4-19
Picture generation with image trans-
formation.

ELSE IF OP < 1 THEN DOSTYLE(OP)

ELSE IF OP = 1 THEN DOMOVE(X, Y)
ELSE IF OP = 2 THEN DOLINE(X, Y)
ELSE DOPOLYGON(OP, X, Y, NTH);
END;
RETURN;
END;

4.12 Algorithm LOAD-POLYGON(I, EDGES) (A modification of algorithm 3.9) A

routine to retrieve polygon side information from the display file
Positions are converted to actual screen coordinates
Arguments I the display-file index of the instruction
EDGES for return of the number of sides stored
Global WIDTH-START, HEIGHT-START starting index of the screen
WIDTH width in pixels of the screen
HEIGHT height in pixels of the actual screen
Local X1, Yl, X2, Y2 edge endpoints in device coordinates
I1 for stepping through the display file
K for stepping through the polygon sides
DUMMY for a dummy argument
BEGIN
set starting point for a side
GET-TRANSFORMED-POINT(I, SIDES, X1, Y1);
X1 <— X1 * WIDTH + WIDTH-START + 0.5;
adjust y coordinate to nearest scan line
Y1 <INT(Y1 * HEIGHT + HEIGHT-START + 0.5);
 TRANSFORMATIONS 131

get index of first side command

Il<I+1;
initialize an index for storing side data
EDGES < 1;
a loop to get information about each side
FOR K = 1 TO SIDES DO
BEGIN
get next vertex
GET-TRANSFORMED-POINT(I1, DUMMY, X2, Y2);
X2 <— X2 * WIDTH + WIDTH-START + 0.5;
Y2 <— INT( Y2 * HEIGHT + HEIGHT-START + 0.5);
see if horizontal line
IF Yl = Y2 THEN X1 < X2
ELSE
BEGIN
save data about side in order of largest y
POLY-INSERT(EDGES, X1, Y1, X2, Y2);
increment index for side data storage
EDGES <- EDGES + 1;
old point is reset
Rye?
XG XO.
END;
<a eal:
END;
set EDGES to be a count of the edges stored.
EDGES <- EDGES -—1;
RETURN;
END;

The GET-TRANSFORMED-POINT routine retrieves an instruction from the dis-

play file and applies the transformation to it.

4.13 Algorithm GET-TRANSFORMED-POINT(NTH, OP, X, Y) Retrieve and trans-

“form the Nth instruction from the display file
Argument NTH the index of the desired instruction
OP, X, Y the instruction to be returned
Global IMAGE-XFORM a 3 X 2 array containing the image transformation
BEGIN
GET-POINT(NTH, OP, X, Y);
IF OP > 0 OR OP < -31 THEN DO-TRANSFORMATION(X, Y, IMAGE-XFORM);
RETURN;
END;

TRANSFORMATIONS AND PATTERNS

In Chapter 3 we described how patterns might be used to fill polygons. We showed
how to implement patterns which are registered to the imaging surface. This is what is
needed for patterns which give gray levels or simple textures such as stripes or weaves.
132 CHAPTER FOUR

But there is an alternative use of patterns where it would be better if the pattern could
be moved with respect to the imaging surface. This occurs when patterns are used to
display pictures. It is sometimes better to represent a picture directly as a pixel pattern,
instead of using lines and polygons. This may be because the shapes in the picture are
curved (for example, characters). Or it may be that there are no lines at all, as often oc-
curs with pictures extracted from photographs. For this type of pattern we may want to
be able to move it around on the display. We might also like to be able to change the
scale and orientation. We might like to rotate the picture or make it bigger. In other
words, we would like to apply a transformation to the pattern. (See Figure 4-20.)
We shall show one way that this can be done using the image transformation
which we have just developed. The resulting program will behave in the following
way. Suppose we create a polygon anda pattern for filling it. Now suppose that we
apply an image transformation to scale the polygon; we shall also scale the pattern so
that the picture looks the same except for size. If we use the image transformation to
rotate the polygon, the pattern will rotate with it, and if we use the image transforma-
tion to translate the polygon, the pattern it is filled with will be translated, too. The
way we shall implement this is by extending the FILLIN algorithm. FILLIN deter-
mines the intensity value for each point in the polygon. In Chapter 3 we showed how
the intensity for a point [x, y] could be found by looking in a pattern table; but now,

FIGURE 4-20
When the filling pattern is a picture, it
should be transformed with the polygon.
 TRANSFORMATIONS 133

FILLIN must find the intensity value for the transformed point [x’, y’] = [x, y]H,
where H is the image transformation and [x, y] is in normalized device coordinates.
Given the point [x’, y’], we need to find the point [x, y] and use it to look up the inten-
sity in the pattern table. In effect, we are asking for the intensity of the original pattern
at the point which gets transformed into the point being imaged. To find [x, y] from
[x’, y’], we need the inverse transformation. We can build the inverse transformation
from the inverses of the scale, rotation, and translation components.
The inverse of the scaling transformation of Equation 4.37 is

& 0 0
st=|/0 £ 0 (4.57)
OU ol
For example, scaling by one-half undoes the effect of scaling by two.
The inverse of the rotation matrix of Equation 4.39 is

cos@ -sing 0 cos(-@) sin(-@) 0O

R‘'=|sing@ cos@ O|=|-sin(-@) cos(-69) 0 (4.58)
0 0 1 0 0 1

This says that we can undo the effect of a counterclockwise rotation by a rotation of the
same amount in the clockwise direction.
Finally, the inverse of the translation matrix of Equation 4.41 is

t—- 0-6
Peat 0-0 (4.59)
mty = fy" 1

We undo the effect of a translation by translating the same amount in the opposite di-
rection.
Our image transformation is the product of a scale, a rotation, and a translation.
Its inverse can be built from the inverses of its components. However, when we take
something apart, we do it in the reverse sequence of when we put it together, and the
inverse of the image transformation H is the product of its components in reverse
order.

Ho (eR) = 7 Rel (4.60)

This is almost what we need for transforming the pattern. The image transformation is
applied to normalized device coordinates, but the points in FILLIN have been trans-
formed to the actual device coordinates; so in order to apply H™', we should first con-
vert the point back to the normalized coordinates by applying the inverse of this coordi-
nate transformation. Once we have applied the inverse of the image transformation,
we must convert the result back again to the actual device coordinates. This may seem
like a lot of work, but actually it is not, because the coordinate transformation matrix
(and its inverse) can be multiplied in with the inverse image transformation matrix so
that there is only a single net transformation which we must apply to each point.
134 = cHaAPTER FOUR

We saw the transformation from normalized device coordinates to actual device

coordinates in Equation 4.43. This is the product of a scale (S,) by the WIDTH and
HEIGHT, followed byatranslation (T,,) by WIDTH-START and HEIGHT-START. So
the full transformation which should be applied to each"point of the polygon being
filled is
Tos he oe re
Let’s extend the BUILD-TRANSFORMATION algorithm so that it builds both
the image transformation and the inverse transformation needed to transform patterns.

4.14 Algorithm BUILD-TRANSFORMATION (Revision of algorithm 4.9) Routine

to build the image transformation matrix and its inverse
Global ANGL, SCLX, SCLY, TRNX, TRNY the transformation parameters
IMAGE-XFORM a 3 2 array containing the image transformation
INVERSE-IMAGE-XFORM a 3 x 2array for the inverse of the image
transformation
BEGIN
IDENTIT Y-MATRIX(IMAGE-XFORM);
MULTIPLY-IN-SCALE(SCLX, SCLY, IMAGE-XFORM);
MULTIPLY-IN-ROTATION(ANGL, IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(TRNX, TRNY, IMAGE-XFORM);
IDENTIT Y-MATRIX(INVERSE-IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(— WIDTH-START, — HEIGHT-START,
INVERSE-IMAGE-XFORM);
MULTIPLY-IN-SCALE(1 / WIDTH, | / HEIGHT, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(— TRNX, — TRNY, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-ROTATION( — ANGL, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-SCALE(1 / SCLX, 1 / SCLY, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-SCALE(WIDTH, HEIGHT, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(WIDTH-START, HEIGHT-START,
INVERSE-IMAGE-XFORM);
RETURN;
END;

Before presenting the modified FILLIN algorithm, let’s define a flag which the
user can Set to indicate whether or not he wants patterns to be transformed.

4.15 Algorithm SET-TRANSFORM-PATTERN(ON-OFF) User routine to indicate

whether polygon fill patterns should undergo the image transformation .
Argument ON-OFF the user’s choice
Global XFORM-PATTERN a flag to indicate transformation of patterns
BEGIN
XFORM-PATTERN < ON-OFF;
RETURN;
END;

Now we shall revise the FILLIN algorithm to allow image transformations on the
patterns used to fill polygons.
 TRANSFORMATIONS 135

4.16 Algorithm FILLIN(X1, X2, Y) (Revision of algorithm 3.17) Fills in scan line Y
from X1 to X2
Arguments X1, X2 end positions of the scan line to be filled
Y the scan line to be filled
Global FILL-PATTERN pattern table index of the pattern to use
PATTERN-X, PATTERN-Y, PATTERNS the pattern table
FRAME the two-dimensional frame buffer array
XFORM-PATTERN a flag to indicate transformation of patterns
INVERSE-IMAGE-XFORM a 3 2 array for the inverse of the image
transformation
Local X for stepping across the scan line
PX, PY for accessing the pattern
PATTERN-TO-USE the pattern to be used in filling
PAT-X, PAT-Y the x and y dimensions of the pattern
BEGIN
IF X1 = X2 THEN RETURN;
PAT-X <— PATTERN-X[FILL-PATTERN];
PAT-Y <— PATTERN-Y[FILL-PATTERN];
if a local array is used as shown in the following statement,
then the implementation should avoid copying of individual elements
PATTERN-TO-USE < PATTERNS[FILL-PATTERN];
IF XFORM-PATTERN THEN
BEGIN
FOR X = X1 TO X2 DO
BEGIN
PX — X;
PY <-Y;
DO-TRANSFORMATION(PX, PY, INVERSE-IMAGE-XFORM);
PX <- MOD(INT(PX + 0.5), PAT-X) + 1;
PY — MOD(INT(PY + 0.5), PAT-Y) + 1;
FRAME[X, Y] < PATTERN-TO-USE[PX, PY];
END;
END
ELSE
r BEGIN
PX <— MOD(X1, PAT-X) + 1;
PY — MOD(Y, PAT-Y) + 1;
FOR X = X1 TO X2 DO
BEGIN
FRAME[X, Y] < PATTERN-TO-USE[PX, PY];
IF PX = PAT-X THEN PX < 1
ELSE PX <— PX + 1;
END;
END;
RETURN;
END;

Note that for pattern sizes restricted to powers of 2, the MOD operation can be re-
placed bya logical AND instruction, which is usually much faster.
136 = CHAPTER FOUR

The algorithm has been extended to include a check for transformation of the
pattern. If the transformation should be done, then the inverse image transformation is
applied to each point that is to be filled. The result is rounded to the nearest pixel, and
MOD operations are used to determine the appropriate pattern element. Although they
may look different, this new portion of the algorithm and the portion written in Chap-
ter 3 are really quite similar. In fact, if the transformation is the identity, then the top
and bottom halves of the algorithm have exactly the same effect. The difference is that
in the untransformed case we can keep track of our position in the pattern, and can
thereby avoid doing the MOD operations for every point. In the transformed case we
cannot do this, and the MOD operators are needed.
Transformations on patterns should be used for pictures, but should be avoided
for simple shading patterns and repetitive designs. There are two reasons for this. One
is that the transformation of every pixel in the filled area can be quite time-consuming.
The second is that transforming a pattern introduces aliasing effects. When trans-
formed, the pixel positions do not correspond directly to the pattern positions. We must
guess the proper pixel intensity from the intensity value of a nearby pattern element.
This guess introduces errors. Antialiasing techniques can be used at the expense of
more time. For example, we can set the pixel intensity to be a weighted average of the
nearby pattern element values.

INITIALIZATION
We need a routine which will initialize the parameters so that the user will not have to
set them unless he wishes to. The default parameter values should give the identity
transformation.

4.17 Algorithm INITIALIZE-4 Initialization routine

BEGIN
INITIALIZE-3;
CALL SCALE(1, 1);
CALL ROTATE(0);
CALL TRANSLATE(0, 0);
SET-TRANSFORM-PATTERN(FALSB);
RETURN;
END;

DISPLAY PROCEDURES
There may be times, when the image transformation presented here is inadequate,
when the user may wish to perform more than just a single scale, rotation, and transla-
tion operation. For example, a rotation about an arbitrary center point is most easily
handled by a translation, followed by a rotation, followed by another translation. In
these situations, the user should add one or more transformation levels
to the system.
He should write his own LINE and MOVE routines which multiply their
arguments by
his own transformation matrix before calling the system LINE-ABS-2 and
MOVE-
 TRANSFORMATIONS 137

ABS-2 routines to actually do the drawing. A transformation matrix designed and

created by the user may be as complex as necessary, involving many component trans-
formations. Routines written according to this prescription will transform point values
before they are entered into the display file. The system extension detailed in the above
algorithms, on the other hand, operates on points as they are read from the display file.
Transformations may be carried out at both stages of the processing.
One situation in which multiple transformations are useful occurs when a picture
is made of a few basic components combined according to some hierarchical structure.
For example, we may have a routine which draws a flower petal. By combining petals
with different positions and orientations, flowers may be drawn. By combining flowers
with different sizes and positions, a flower bush may be created. Transformations of
several bushes can form a garden, and so on. (See Figure 4-21.)
Pictures with this structure lend themselves to a subprogram organization. A pro-
gram to draw a garden could do so by several calls on a subprogram which draws a
flower bush. The subprogram which draws the flower bush could do so by several calls
on a flower-drawing subprogram. The flower-drawing subprogram could use several
calls on the petal subprogram. Notice, however, that these subprogram calls are a little
more complicated than those of normal programming languages; they involve the es-
tablishment of a transformation matrix. An ordinary subprogram call (CALL PETAL)
would always produce the image with the same size and orientation. What is needed is
a call which sets up a transformation which is applied to all the points generated in the
subprogram before they are entered into the display file, for example,
CALL PETAL WITH SIZE(SX, SY), ANGLE(A), TRANSLATION(TX, TY).
These calls, which involve establishment of a transformation, are named display
procedure calls, and the subprograms which draw subpictures are known as display pro-
cedures. Because display procedure calls can be nested, there can be multiple transfor-

Cay Flower

SB BS &
Garden
Bush
FIGURE 4-21
Hierarchical picture structure.
138 cHapTER FOUR

mations. Suppose, for example, that the GARDEN program calls the FLOWER-BUSH
display procedure with transformation T,. FLOWER-BUSH calls the FLOWER display
procedure with transformation T. Finally, FLOWER calls the PETAL display proce-
dure with transformation T;. Then, the overall transformation applied to points gener-
ated in PETAL would be the product T,T,T,. We can see that each time a new display
procedure call is executed, the overall transformation matrix is multiplied on the left
by the transformation for that display procedure. Here is a case where pre-multiplica-
tion is appropriate. In the GARDEN procedure, we have the identity matrix I. When
FLOWER-BUSH is called, this is multiplied by T, to give T,. When this, in turn, calls
FLOWER, the T, transformation is multiplied in to give T,T,. Finally, when PETAL is
called, the third transformation is also present: T;T,T,. What happens when PETAL
finishes and control is returned to FLOWER? We would expect to have the overall
transformation which is appropriate for FLOWER, namely T, T,. Likewise, when con-
trol is returned to FLOWER-BUSH from FLOWER, the transformation should be T,,
and when control returns to GARDEN, the overall transformation should once again be
the identity I. There must therefore be some way of saving the overall transformation
before multiplying by the additional transformation for a new display procedure call,
so that it may be restored when control returns from that display procedure. When
FLOWER calls PETAL, it saves the current overall transformation T,T,, multiplies it
by T3 to get the new overall transformation T3;T,T,, and transfers control to the
PETAL. When PETAL is finished and ready to return, it first restores the overall trans-
formation to the value that was saved, T,T,, and then returns control to FLOWER.
Each display procedure call must save the current overall transformation matrix. Be-
cause there can be nested calls, several transformation matrices may have to be stored
simultaneously. One possible data structure for storing these matrices is a stack. The
last item stored in a stack is the first item to be removed, so it matches the last-en-
tered—first-returned nature of subroutines.
In summary, then, a display procedure call involves the following:

1. Saving the overall transformation matrix

2. Multiplying the overall transformation matrix on the left by the transformation in
the call to form a new overall transformation matrix
3. Transferring control to the display procedure

A return from a display procedure involves the following:

1. Restoring the overall transformation matrix from the value saved

2. Returning control to the calling program

The user’s LINE and MOVE commands within the body of a display procedure should
do the following:

1. Multiply point coordinates by the current overall transformation matrix to

get the
transformed point.
 TRANSFORMATIONS 139

2. Enter the transformed values into the display file via the system LINE-ABS-2 or
MOVE-ABS-2 commands.

AN APPLICATION
Let’s consider how our graphics system might be useful in producing animated films.
Animation is done by photographing a sequence of drawings, each slightly different
from the previous. For example, to show a person moving his arm, a series of draw-
ings is photographed, each drawing showing the arm at a different position. When the
images are displayed one after another by a movie projector, we perceive the arm as
moving through the sequence. (See Figure 4-22.)
From the artist’s point of view, it is desirable to have only a small amount of mo-
tion occurring at any one time. The picture can then be constructed of a background
which does not change and a foreground which alters from frame to frame. The fore-
ground may be drawn upon a piece of clear plastic and then overlaid upon the back-
ground. If the portion of the image which changes is small, then the foreground image
which must be redrawn for each frame is small and will require less work. What may
be difficult for the human artist is changing the entire scene. For example, if one
wished to give the impression of moving into the scene, one might scale the back-
ground, making each frame slightly larger. While this sort of change is difficult for
the human artist, it is easy for the computer. It may therefore be beneficial to have
the computer generate the background scene, while the human artist superimposes the
foreground action.
Let’s see how our graphics system could be used to generate the background.
First, we must construct the full background drawing by using the LINE and POLY-
GON commands of the previous chapters. Let us assume that this has been done and
that we have a routine which will enter appropriate instructions into the display file.
By repeatedly executing these commands, the background images for each photograph
may be generated. Suppose that the scene is a city street. To show a character moving
down the street, we may really want to keep the character centered and move the scen-
ery past him. This can be done by shifting our background image with the TRANS-
LATE(TX, TY) command. Before each display, we add a little more to TX. Each
frame will show the background shifted a little more to the right. This will give the im-
pression that our character has moved to the left. (See Figure 4-23.)
Or suppose that we wish to show our character approaching a building. We could
give this effect by having the building grow larger. It could be done by using the
SCALE(SX, SY) routine to make the building grow, and the TRANSLATE(TX, TY)

BeA
FIGURE 4-22
Animation of arm movement byaseries of pictures, each slightly different from the previous.
140 CHAPTER FOUR

t FIGURE 4-23
Apparent movement by translation of the background.

routine to keep it centered. With each frame, SX and SY are increased slightly to make
the building appear alittle larger. (See Figure 4-24.)

FURTHER READING
Other techniques for transforming patterns are found in [BRA80]. Transforming im-
ages is also considered in [CAT80]. Aliasing can be a major problem for transformed
patterns. One approach to the problem is to find all pattern elements which transform
to part of the displayed pixel and then to average their intensity values. A clever way to
find the sum of the pattern element intensities is to store the progressive sums of inten-
sity values in the pattern table so that the total intensity for a rectangular area is given
by the difference in the values at the corners [CRO84]. A formal description of trans-

25d
Ox
+O FIGURE 4-24
Apparent motion by scaling of the background.
 TRANSFORMATIONS 141

formations and clipping in a hierarchical picture structure is described in [MAL78].

Display procedures are described in [NEW71].

[AHU68] Ahuja, D. V., and Coons, S. A., “‘Geometry for Construction and Display,’’ JBM Systems Journal,
vol. 7, no. 3/4, pp. 188-205 (1968).
(BRA80] Braccini, C., Marino. G., ‘Fast Geometrical Manipulations of Digital Images,’’ Computer
Graphics and Image Processing, vol. 13, no. 2, pp. 127-141 (1980).
(CAT80] Catmull, E., Smith, A. R., ‘‘3-D Transformations of Images in Scanline Order,’’ Computer
Graphics, vol. 14, no. 3, pp. 279-285 (1980).
[CRO84] Crow, F. C., ‘‘Summed-Area Tables for Texture Mapping,’’ Computer Graphics, vol.18, no. 3, pp.
207-212 (1984).
[MAL78] Mallgren, W. R., Shaw, A. C., ‘‘Graphical Transformations and Hierarchic Picture Structures,”’
Computer Graphics and Image Processing, vol. 8, no. 3, pp. 237-258 (1978).
[NEW71] Newman, W. M., ‘‘Display Procedures,’’ Communications of the ACM, vol. 14, no. 10, pp. 651-
660 (1971).

EXERCISES
4-1 Perform the following indicated matrix multiplications:

r’o|*)2 4
a)|9 S| i |
On| 1 2
»|6 | E i
Gshl lad?
c)) 4 4 |} i

a|2 4| |3 4
of3S52
a [4 + 20
piparbasr 3 4 1|
Ube seed t
|-2 3 0

: 000
eee |0
ylo o 1] 0 0

4°00
h)|1 1 1 040
001
01 0
{1 0 1] |-1 0 0
001
2. 4. DL Aan
west oT tt 8
FO 1 |< o. 2 a
142 cHAPTER FOUR

4-2 Write a2 X 2 transformation matrix for each of the following scaling transformations:
a) The entire picture three times as large
b) The entire picture one-third as large
c) The x direction four times as large, the y direction unchanged
d) The y lengths reduced to two-thirds their original value, the x lengths unchanged
e) The x direction reduced to three-fourths the original, y direction increased by a factor
of seven-fifths
4-3 Give 3 X 3 homogeneous-coordinate transformation matrices for each of the scaling trans-
formations given in Exercise 4-2.
4-4 Write a2 X 2 transformation matrix for each of the following rotations about the origin:
a) Counterclockwise by 7
b) Counterclockwise by 7/2
c) Clockwise by 11/2
d) Counterclockwise by 7/4
e) Counterclockwise by 57/4
4-5 Give a 3 X 3 homogeneous-coordinate transformation matrix for each of the rotations of
Exercise 4-4.
4-6 Give a 3 X 3 homogeneous-coordinate transformation matrix for each of the following
translations:
a) Shift the image to the right 3 units.
b) Shift the image up 2 units.
c) Move the image down !/2 unit and right 1 unit.
d) Move the image down 2/3 unit and left 4 units.
4-7 Give a single 3 X 3 homogeneous coordinate transformation matrix which will have the
same effect as each of the following transformation sequences:
a) Scale the image to be twice as large and then translate it 1 unit to the left.
b) Scale the x direction to be one-half as large and then rotate counterclockwise by 1/2
about the origin.
c) Rotate counterclockwise about the origin by 7/2 and then scale the x direction to be
one-half as large.
d) Translate down !/2 unit, right '/2 unit, and then rotate counterclockwise by 17/4.
e) Scale the y coordinate to make the image twice as tall, shift it down 1 unit, and then
rotate clockwise by 17/6.
4-8 Show how reflections in the line y = x and in the line y = — x can be performed by ascal-
ing operation followed by a rotation. (Scaling by negative factors is permitted.)
4-9 Suppose that a figure is centered at (0.5, 0.5):
a) Give the calls to SCALE, ROTATE, and TRANSLATE which will show the figure
scaled twice as large and still centered on the screen.
b) Give the calls to SCALE, ROTATE, and TRANSLATE which scale the figure bya fac-
tor S and keep it centered on the screen.
c) Give the calls to SCALE, ROTATE, and TRANSLATE which will rotate the figure
counterclockwise by an angle A and still keep it centered on the screen.
d) Give the calls to SCALE, ROTATE, and TRANSLATE which will show the figure re-
flected in the line x = 0.5.
e) Give the calls to SCALE, ROTATE, and TRANSLATE which will show the figure re-
flected in the point (0.5, 0.5); that is, x; = 1 — xandy, = 1 — y.
 TRANSFORMATIONS 143

4-10 a) Prove that two scaling transformations commute; that is, S,S, = S,S).
b) Prove that two two-dimensional rotations about the origin commute; R;R, = RR).
c) Show that two-dimensional scales and rotations do not commute in general. Under
what restriction on scaling transformations is commutation of scales and rotations guaranteed?
4-11 Show how shear transformations may be expressed in terms of rotations and scales. Show
how rotations can be expressed in terms of shears and scales. What scaling operations can be ex-
pressed as shears?
4-12 a) Show that a rotation about the origin can be done by performing three shearing trans-
formations (with no scaling). This can be a fast way to rotate bitmaps since shears can often be
carried out by moving entire blocks of pixels.
b) On each shearing operation, a quantization error in pixel position can be made. What
is the maximum cumulative error in pixel position for a rotation conducted as three shears?

of8 3] [8 8] ol
4-13 Describe the visual effect on an image of each of the following transformations:

1 0
Zima aoa o(8 t|

oS3][fa] oft3
4-14 Find the inverse of each of the matrices in Exercise 4-13, if it exists.
4-15 Express each of the transformations in the OTHER TRANSFORMATIONS section as a
product of rotations and scales. Also give the inverse transformation.

PROGRAMMING PROBLEMS
4-1 Implement algorithms 4.1 through 4.13, and 4.17, which extend the graphics system to in-
clude image transformations.
4-2 Write a program which draws a house-shaped, five-sided polygon. Show this figure scaled,
rotated, translated, and under a combination of all three transformations. Show the figure both
filled and unfilled.
4-3 Write the three routines
INQUIRE-SCALE(SX, SY)
INQUIRE-ROTATE(A)
INQUIRE-TRANSLATE(TX, TY)
which will return the current image transformation parameter settings.
to
4-4 Eye chart program: Usingafilled polygon to construct a square letter E, write a program
of
successively show this letter at different sizes and orientations. The orientation should be one
the four E,3,W,m and should be picked at random for each image. The size should begin
(See
large, but get smaller with each frame. The letter should always be centered on the screen.
page 144.)
by perform-
4-5 Write a procedure PIE(A1, A2, R) which draws a pie-shaped segment of a circle
the result of the WEDGE procedure of Programming Problem
ing an image transformation on
have radius R,
3-12. The segment of a circle should have its point centered at (0.5, 0.5), should
and should lie between angles A, and A>.
 OR ee
KEEOK
AEE ETE TREE EEE EEE OH
REE
ROR EE ee
ROE REE KERR EEE ERAT K EERE HH
RR SR RR Re RR i aieee ee Ee ec aieieaiae a
RRR EEK EH
KCeRe kee KE RH
eee
SORE KER EH
eR RRR EKER EH
REREERERERERE EE EEEEREETT
REET ERE EEA ETAT AEST EES EE
Re Re RR NRK RRR RC RC a a
CREE REE TEE ee
CEE ERRATA AK ee ee
eR eR eR a a CC a
REE EE RR EA REE EEE RE EAE EEO AKAK ERE R EEE
CREEK EE RH
RRR RK
ROKR HE
SKE KEE
ehoee
SRE EARS EASTER ETE RETESET KEE TKR EES
RECERCAT EERSTE RSE RE TH ES
EEE EAE EAE ECR RETR E TE
CREE ETT EATER EKER ERS
AREER EAE AERA RE
CAESAREA HOKE Ke
KERR A RE RAAAE EE RAEA EERE S ee

CREE
RRR EEE THEE EEE S
ERK a
CRETE RETA TEER EERE RTS
EERE AREA ETA ETRE TERETE KER EE
RHR RC RCCC Rai a Cea ek a
Seeee eases
CC KER SHEE
keke ee eee
cteeeeaeee
RETREAT Se
RRORREREEER EKER ERS AREA EHR HERR E ERROR ERE ES
CREEK EERE EE EERE KEES ER ARR RRR ER EE
KERR ee
REE AEEAAAA SKS AKRR RR ETRE SOSe
EKER EATER TREK ERR SER ERS
re eee
eee keene ee
Stee eeoae
RK ROE
KEKE RKEE EERSTE ATER KKK TREKS SSS STE
REAR A EAA EEE ERS e
TKO EASES TERE RRS
EKER EAE R RE TEE
KERR EKKAREE REE RRR EET RS
CERES ERE |

CRAKEAR Oe
CREE RS
FREER ERE KEE REE REE ERE RRR EREEE e ee
CRA ARERR KE Re
SEEKER EAA RETA AAR EERE RATERS ee ee eeeeense eeecese eeeteese
Stk eee etna eaxeee eth ceeene Reece eee eeeeeee eeeeees
eR KR RH Km eee eek ee ce ek eee ce eeeen
eee eee e see eeeeses eee eee eee eeae eeteees aeeetes
SRK RH Kee S ERE eee eee oe ere eeee keeeeee eeeeeee
et ee eo KEE tee nee Sekeeeee eerases
See R RSH TCE K Eee Fee RE eE eee ease eeeeeee eteeee
TTR RE RH eek ee ee che eesonm ceeeek eee tekeeeee aeeeene
KEKE Re eR Oo eke Re Ree eeeeeee ceeeeae steeeee
KeRKR ERED cKkk eee HH TKR ETOH et eene eeetese setaeee
Khe RR ee eee eee cee eK ete tease eee eeee
eee ee Ree kkk eee ettekeee eeeeeee eettene
kee KK Ktekeeeae kkk eee eee eeee eteeece seeenee
eee Re RRR Re
SEEKER
EREEE EERE TRE RE
Re Km ee ke ee eR RRR Re
SERRE KERRERKE TREES EEE E EH
eee eek Kee *eKKKR ERE Ree Teo H SSSR KEKE EERE
ERE REE Ee
ek eR ee ke me
HERRERA
EKER RRR ERR EROS
RET KE Sek eee e ee RECKRE REE PSC SKKEKEEEKE
KERR SEEKER ESEE

144
 TRANSFORMATIONS 145

4-6 Suppose that procedure FOO has LINE, MOVE, and POLYGON drawing commands
for
some image. Write a program to repeat the following steps.
a) Call FOO to create the image.
b) Form a new image transformation which will show the image rotated m/5 coun-
terclockwise and scaled by 0.8 from the last image displayed. The image should remain centered
on the screen.
c) Display the resulting image.

Test this program for five iterations, where FOO draws the square (0, 0), (0, 1), (1, 1), (1, 0).
4-7 Implement algorithms 4.14 through 4.17 for transformation of a polygon-filling pattern. Test
your implementation by showing a polygon filled with a pattern when untransformed, trans-
lated, scaled, rotated, and under a combination of all three transformations. Show it under both
settings of the XFORM-PATTERN flag.

ES

4-8 An application program can make greater use of transformation than we have provided. One
might wish more general transformations such as the reflections and shears, or one might want
to apply long sequences of transformations. Write user procedures TMOVE-ABS-2(X, Y), T-
MOVE-REL-2(DX, DY), T-LINE-ABS-2(X, Y), T-LINE-REL-2(DX, DY), T-POLYGON-ABS-
2(AX, AY, N), and T-POLYGON-REL-2(AX, AY, N), which apply a general transformation to
the coordinates before passing them on the MOVE-ABS-2, MOVE-REL-2, etc. Also write pro-
cedures T-CLEAR, which sets this application transformation to the identity transformation, and
T-MULTI-XFORM(A, B, C, D, E, F), which multiplies the application transformation by
A D0
BE O
Cask oA
Optionally, you might write individual translation, scale, rotation, and shear routines which are
special cases of the TMULT-XFORM. Note that the ‘‘absolute’’ operations have full coordinates
for arguments and are multiplied by the full transformation, but the ‘‘relative’’ operations have
vectors as arguments. Vectors undergo the 2 X 2 scale and rotation transformations because they
have length and direction, but they do not undergo translation because they do not have position.
*4-9 Implement a display procedure mechanism.
 CHAPTER

FIVE
SEGMENTS

INTRODUCTION
In the preceding chapters we have organized our display as a single picture. In reality,
however, the image on the display screen is often composed of several pictures or
items of information. A single image may contain several views of an object. It may
have a picture of the overall object and a close-up of a particular component. It may
also contain information about the object, instructions for the user, and, possibly, error
information. As an example, we may wish to display the design plans for a building.
The plans may contain structural diagrams, electrical diagrams, plumbing diagrams,
and heating diagrams. We might wish to show all this information simultaneously or, at
other times, look at the individual elements. As another example, consider an ani-
mated display of a spaceship in motion. We might want to show the spaceship at differ-
ent positions while the background remains fixed, or we might choose to keep the ship
centered on the screen and move the background. The transformations of the last chap-
ter tell us how to shift the position of the image, but now we wish to apply them to
only a portion of the scene (either the ship or the background, but not both).
We would like to organize our display file to reflect this subpicture structure. We
would like to divide our display file into segments, each segment corresponding to a
component of the overall display. We shall associate a set of attributes with each seg-
ment. One such attribute is visibility. A visible segment will be displayed, but an invisi-
ble segment will not be shown. By varying the settings of the visibility attribute, we
can build a picture out of the selected subpictures. Consider the building plan applica-

146
 SEGMENTS 147

tion. We might, for example, in one display select structural and electrical information
for a building by making the first and second segments visible and the third and fourth
Segments invisible. In another display we could select structural and plumbing infor-
mation by making only the first and third segments visible. While all display informa-
tion may be present in the machine, we can selectively show portions of it by designat-
ing which segments of the display file are to be interpreted.
Another attribute which can be associated with each segment is an image trans-
formation. This will allow the independent scaling, rotating, and translating of each
segment. For the spaceship, we could put the ship in one segment and the background
in a different segment. We could then shift either of them by means of an image trans-
formation, while leaving the other unchanged. (See Figure 5-1.)
In this chapter we shall consider segmentation of the display file. We shall learn
how to create, close, rename, and delete segments. We shall also consider such dis-
play-file attributes as visibility and image transformation.

THE SEGMENT TABLE

We shall begin our discussion by considering some of the information that must be as-
sociated with each segment and how this information might be organized. We must
give each segment its own unique name so that we can specify it. If we are to change
the visibility of a segment, we must have some way to distinguish that segment from
all the others. When we refer to a display-file segment, we must know which display-
file instructions belong to it. This may be determined by knowing where the display-
file instructions for the segment begin and how many of them there are. For each seg-
ment, we shall need some way of associating its display-file position information and
its attribute information with its name. We need to organize this information so that
given the segment name, we can look up or alter its attributes, or given its name, we
can interpret the corresponding display-file instructions. In our system we shall do this

FIGURE 5-1
Transformation of a portion of the display.
148 CHAPTER FIVE

by forming a segment table. We shall use a number for the name of the segment. Sim-
ple arrays will serve to hold segment properties, and the segment name will be used as
the index into these arrays. We shall have one array containing display-file starting lo-
cations. A second array will hold segment size information, while a third will indicate
visibility, and so on. To find out the size of, say, the third segment, we would look for
the third element in the size array. (See Figure 5-2.)
The number | will refer to the first named segment of the table, 2, to the second
named table entry, and so on. We shall have table entries (arrays) for holding the dis-—
play-file starting position, for the segment size, and for attributes such as visibility and
the image transformation parameters. If we wish not to show segment number 3, we
will set the corresponding entry in the visibility array to off. When we show our pic-
ture, we shall consult the segment table to determine which segments are visible. For
each visible segment, we shall look up its starting point and size and pass this informa-
tion along to our display-file interpreter. The interpreter will therefore only interpret
those segments which are visible.
There are other possible schemes for implementing the segment table, many with
substantial advantages over the one which we have chosen here. We have selected this
design because it allows simple accessing. It does not require any new data-structuring
concepts, and its updating is straightforward.
We would like this extension to our graphics system to be compatible with our
previous programs, which did not specify any segments. Perhaps the easiest way to do
this is to say that when no segment name is specified, the instructions will be placed in
a special ‘‘unnamed’’ segment. Information, such as display-file starting position and
segment size, must be stored for the unnamed segment just as it is for named seg-
ments. We shall do this by placing a special entry in the segment table for the unnamed

SEGMENT SEGMENT- SEGMENT-

NAME START SIZE VISIBILITY SCALE-X a
0

eK
FWN
eee

FIGURE 5-2
The segment table.
 SEGMENTS 149

segment. In this implementation, we have chosen to associate the unnamed segment

with table index 0. That is, the segment table arrays will be dimensioned with lower
bound 0 instead of 1. Thus SEGMENT-SIZE[1] will still be the number of instructions
of the segment with name 1, but SEGMENT-SIZE[0] will be the number of instruc-
tions in the unnamed segment.

SEGMENT CREATION
Let us consider the process of creating or opening a segment. When we create a seg-
ment, we are saying that all subsequent LINE, MOVE, TEXT, or POLYGON com-
mands will be members of this segment. We must give the segment a name so that we
can identify it. We might, for example, say that this is segment number 3. Then all fol-
lowing MOVE and LINE commands would belong to segment 3. We could then close
segment 3 and open another segment, say segment 5. The next MOVE and LINE com-
mands would belong to segment 5.
The first thing we do when we create a segment is check to see whether some
other segment is still open. We cannot have two segments open at the same time be-
cause we would not know to which segment we should assign the drawing instruc-
tions. If there is a segment still open, we have an error. If no segment is currently
open, we should check to make sure that we have a valid segment name. If the seg-
ment name is correct, check to see whether there already exists a segment under this
name. If so, we again have an error. We initialize the items in the segment table under
our segment name to indicate that this is a fresh new segment. The first instruction be-
longing to this segment will be located at the next free storage area in the display file.
The current size of the segment is zero since we have not actually entered any instruc-
tions into it yet. The attributes are initialized to those of the unnamed segment, which
provides the default attribute values. Finally we indicate that there is now a segment
open (the one which we just created).

5.1 Algorithm CREATE-SEGMENT(SEGMENT-NAME) User routine to create a

named segment
Argument SEGMENT-NAME the segment name
Global NOW-OPEN the segment currently open
FREE the index of the next free display-file cell
SEGMENT-START, SEGMENT-SIZE, VISIBILITY
ANGLE, SCALE-X, SCALE-Y, TRANSLATE-X, TRANSLATE-Y
arrays that make up the segment table
Constant . NUMBER-OF-SEGMENTS size of the segment table
BEGIN
IF NOW-OPEN > 0 THEN RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
IF SEGMENT-SIZE[SEGMENT-NAME] > 0 THEN
RETURN ERROR ‘SEGMENT ALREADY EXISTS’;
SEGMENT-START[SEGMENT-NAME] < FREE;
150 CHAPTER FIVE

SEGMENT-SIZE[SEGMENT-NAME] < 0;
VISIBILITY [SEGMENT-NAME] < VISIBILITY [0];
ANGLE[SEGMENT-NAME] < ANGLE[0];
SCALE-X[SEGMENT-NAME] < SCALE-X[0];
SCALE-Y[SEGMENT-NAME] < SCALE-Y([0];
TRANSLATE-X[SEGMENT-NAME] < TRANSLATE-A[0];
TRANSLATE-Y[SEGMENT-NAME] < TRANSLATE-Y(0];
NOW-OPEN <- SEGMENT-NAME;
RETURN;
END;

CLOSING A SEGMENT
With the segment open we can proceed with the entry of display-file instructions, as
we did in previous chapters. Now, however, all commands which are entered are as-
sociated with the open segment. Once we have completed the drawing instructions for
the segment, we should close it. At this point, all that is necessary in closing the seg-
ment is to change the value of the NOW-OPEN variable, which indicates the name of
the currently open segment. We must change it to something, so we shall set it to 0 and
make the unnamed segment the one which is open. In our algorithm below we do this
by placing a 0 in the NOW-OPEN variable. We don’t want to have two unnamed seg-
ments around because we shall only show one of them and the other would just waste
storage. So we delete any unnamed segment instructions which may have been saved.
We initialize the unnamed segment to have no instructions, but to be ready to receive
them in the next free display-file location.

5.2 Algorithm CLOSE-SEGMENT User routine to close the currently open segment
Global NOW-OPEN the name of the currently open segment
FREE the index of the next free display-file cell
SEGMENT-START, SEGMENT-SIZE start and size of the segments
BEGIN
IF NOW-OPEN = 0 THEN RETURN ERROR ‘NO SEGMENT OPEN’:
DELETE-SEGMENT(0);
SEGMENT-START(0] < FREE;
SEGMENT-SIZE[0] < 0;
NOW-OPEN < 0;
RETURN;
END;

DELETING A SEGMENT
The most complex algorithm for this chapter is that for deleting a segment. If a seg-
ment is no longer needed, we would like to recover the display-file storage occupied
by its instructions. Since these instructions will never again be executed, we would
prefer to use this storage for some other segment. At the same time, we would rather
not have to destroy and re-form the entire display file. We may be altering some small
 sEGMENTS 151

portion of an image which is very costly to generate. We would like to be able to delete
(and perhaps re-create with modifications) just one segment, while preserving the rest
of the display file. The method of doing this depends upon the data structure used for
the display file. We have used arrays. Recovery of a block of storage in an array is both
simple and straightforward, but it is not as efficient as some other storage techniques.
What we shall do is take all display-file instructions entered after the segment which
we are deleting was closed, and move them up in the display file so that they lie on top
of the deleted segment. Thus we fill in the gap left by the deleted block and recover an
equivalent amount of storage at the end of the display file. (See Figure 5-3.)
We begin our segment deletion algorithm by checking to make sure that we have
a valid segment name. If the name is correct, we next check that it is not open. Open
segments are still in use; if an attempt is made to delete an open segment, we shall
treat it as an error. We next check to see whether the size of the segment is 0. A seg-
ment with no instructions has nothing to remove from the display file, so no further
processing is necessary. We can now shift the instructions in the display file. We wish
to place instructions on top of the deleted segment. So our first relocated instruction
will be put on top of the first instruction of our deleted segment. We will get the in-
struction to be moved from the first location beyond the deleted segment. We shall
move through the display file, getting instructions and shifting them down until we
come to an unused display-file location. When we have completed moving the display-
file instructions, we can reset the index of the next free instruction to reflect the recov-
ered storage. Finally, we must adjust not only the display file but also the segment
table, since the starting positions of segments created after the segment which was de-
leted will now have changed. We can do this by scanning the segment table and noting
any segments whose starting position lies beyond the starting position of the segment
which we deleted. We change the starting positions by subtracting the size of the de-
leted segment. We can set the size of the segment which we deleted to be 0to indicate
that this segment no longer exists. If we delete a visible segment, a NEW-FRAME ac-
tion is required. These steps are detailed in the following algorithm:

‘wer Segment 3
a
Segment 2 Lae ry Besnents
es re

Segment 3 4 Ss
pea

Segmen t4 be cae

Unused
Unused

FIGURE 5-3
Before After Deleting segment 2 from the display file.
152 CHAPTER FIVE

5.3 Algorithm DELETE-SEGMENT(SEGMENT-NAME) User routine to delete a

segment
Argument SEGMENT-NAME the segment name
Global NOW-OPEN the currently open segment
FREE the index of the next free display-file cell
DF-OP, DF-X, DF-Y the display-file arrays
SEGMENT-START, SEGMENT-SIZE, VISIBILITY part of the segment
table arrays
Constant . NUMBER-OF-SEGMENTS the size of the segment table
Local GET the location of an instruction to be moved
PUT the location to which an instruction should be moved
SIZE the size of the deleted segment
I a variable for stepping through the segment table
BEGIN
IF SEGMENT-NAME < 0 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
IF SEGMENT-NAME = NOW-OPEN AND SEGMENT-NAME # 0 THEN
RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-SIZE[SEGMENT-NAME] = 0 THEN RETURN;
PUT <— SEGMENT-START[SEGMENT-NAME];
SIZE <— SEGMENT-SIZE[SEGMENT-NAME];
GET <— PUT + SIZE;
shift the display-file elements
WHILE GET < FREE DO
BEGIN
DF-OP[PUT] <— DF-OP[GET];
DF-X[PUT] <— DF-X[GET];
DF-Y [PUT] <- DF-Y[GET];
PUT — PUT 1:
GET < GET + 1;
END;
recover the deleted storage
FREE < PUT;
update the segment table
FOR I = 0 TO NUMBER-OF-SEGMENTS DO
IF SEGMENT-START[I] > SEGMENT-START[SEGMENT-NAME] THEN
SEGMENT-START[I] <— SEGMENT-START[I] — SIZE;
SEGMENT-SIZE[SEGMENT-NAME] < 0; :
IF VISIBILITY [SEGMENT-NAME] THEN NEW-FRAME;
RETURN;
END;

We also shall have a routine to delete all of the segments. This will be useful to
the user should he wish to start a completely new picture and will also be useful for
ini-
tialization. One way of doing this is to delete all segments individually. This approach
is independent of the data structure used for the display file. A somewhat more
effi-
cient approach for our particular array structure is to simply set the size value
of all
 SEGMENTS 153

segments to 0 and initialize the free cell index FREE to be the first cell in the
display
file. We also set all starting positions to 1 so that after initialization there will not be
any garbage in these locations which might upset the DELETE-SEGMENT routine.

5.4 Algorithm DELETE-ALL-SEGMENTS User routine to delete all segments

Global NOW-OPEN the segment currently open
FREE the index of the next available display-file cell
SEGMENT-SIZE the segment size array
SEGMENT-START the segment starting index array
Constant © NUMBER-OF-SEGMENTS the size of the segment table
Local I a variable for stepping through the segment table
BEGIN
FOR I = 0 TO NUMBER-OF-SEGMENTS DO
BEGIN
SEGMENT-START[I] < 1;
SEGMENT-SIZE[I] <— 0;
END;
NOW-OPEN < 0;
FREE <1;
NEW-FRAME;
RETURN;
END;

RENAMING A SEGMENT
Another routine which is easy to implement and often useful is renaming a segment.
As an example of how it might be used, consider a display device with an independent
display processor. The display processor is continuously reading the display file and
showing its current contents. (We would not need a MAKE-PICTURE-CURRENT
routine for such a device because the picture is always current.) Now suppose we
wish to use this device to show an animated character moving on the display. This
would be done by presenting a sequence of images, each with a slightly different draw-
ing of the character. Assume we have a segment for the character. Then, for each new
image, we could delete the segment, re-create it with the altered character, and show
the result. The problem with this is that during the time after the first image is deleted
but before the second image is completed, only a partially completed character can be
seen. Since we may begin working on the next image as soon as the last one is com-
pleted, we may in fact be continually looking at only partially completed characters.
To avoid this problem we should not delete a segment until a replacement for it is com-
pleted. This means that both segments must exist in the display file at the same time.
We do this by building the new invisible image under some temporary segment name.
When it is completed, we can delete the original image, make the replacement image
visible, and rename the new segment to become the old segment. These steps can be
repeated to achieve apparent motion. The idea of maintaining two images, one to show
and one to build or alter, is called double buffering. The renaming is carried out by our
RENAME-SEGMENT algorithm. The algorithm checks that the segment names are
154 CHAPTER FIVE

valid and that they are not still open. It also checks against using the name of an al-
ready existing segment. If these conditions are met, the segment table entries for the
old name are copied into the new name position and the size of the old segment is set
back to 0.

5.5 Algorithm RENAME-SEGMENT(SEGMENT-NAME-OLD,

SEGMENT-NAME-NEW) User routine to rename SEGMENT-NAME-OLD to be
SEGMENT-NAME-NEW
Arguments SEGMENT-NAME-OLD old name of segment
SEGMENT-NAME-NEW new name of segment
Global SEGMENT-START, SEGMENT-SIZE, VISIBILITY, ANGLE, SCALE-X,
SCALE-Y, TRANSLATE-X, TRANSLATE-Y the segment table arrays
Constant NUMBER-OF-SEGMENTS the size of the segment table
BEGIN
IF SEGMENT-NAME-OLD < 1 OR SEGMENT-NAME-NEW < 1 OR
SEGMENT-NAME-OLD > NUMBER-OF-SEGMENTS OR
SEGMENT-NAME-NEW > NUMBER-OF-SEGMENTS THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
IF SEGMENT-NAME-OLD = NOW-OPEN OR
SEGMENT-NAME-NEW = NOW-OPEN THEN
RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-SIZE[SEGMENT-NAME-NEW] + 0 THEN
RETURN ERROR ‘SEGMENT ALREADY EXISTS’;
copy the old segment table entry into the new position
SEGMENT-START[SEGMENT-NAME-NEW]
< SEGMENT-START[SEGMENT-NAME-OLD];
SEGMENT-SIZE[SEGMENT-NAME-NEW]
<— SEGMENT-SIZE[SEGMENT-NAME-OLD};
VISIBILITY [SEGMENT-NAME-NEW] < VISIBILITY[SEGMENT-NAME-OLD];
ANGLE[SEGMENT-NAME-NEW] <- ANGLE[SEGMENT-NAME-OLD];
SCALE-X[SEGMENT-NAME-NEW] <— SCALE-X[SEGMENT-NAME-OLD};
SCALE-Y [SEGMENT-NAME-NEW] < SCALE-Y[SEGMENT-NAME-OLD};
TRANSLATE-X[SEGMENT-NAME-NEW]
<— TRANSLATE-X[SEGMENT-NAME-OLD];
TRANSLATE-Y[SEGMENT-NAME-NEW]
<— TRANSLATE-Y[SEGMENT-NAME-OLD};
delete the old segment
SEGMENT-SIZE[SEGMENT-NAME-OLD] < 0;
RETURN;
END;

VISIBILITY
We have talked about the property of visibility. Each segment is given a visibility attri-
bute. The segment’s visibility is stored in an array as part of the segment table. We are-
therefore able to look up the value of each segment’s visibility. By checking this
array
we can determine whether or not the segment should be displayed. We should give
the
user some method of changing the value of a segment’s visibility, so that he can choose
to show or not to show the segment. The following algorithm does this, while freeing
 sEGMENTS 155

the user from any concern about global variables and segment table representation. If
the visibility is being turned off, then a new-frame action is needed. (See Figure 5-4.)

5.6 Algorithm SET-VISIBILITY(SEGMENT-NAME, ON-OFF) User routine to set

the visibility attribute
Arguments SEGMENT-NAME the name of the segment
ON-OFF the new visibility setting
Global VISIBILITY the array of visibility flags
Constant ©NUMBER-OF-SEGMENTS the size of the segment table
BEGIN
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
VISIBILITY[SEGMENT-NAME] < ON-OFF;
IF NOT ON-OFF THEN NEW-FRAME;
RETURN;
END;

IMAGE TRANSFORMATION
An image transformation is carried out on the contents of the display file. If we think
of the display file as containing the picture which we have constructed, then the image

Visibility on

Visibility off

FIGURE 5-4
Visibility on Changing the visibility attribute.
156 CHAPTER FIVE

transformation provides some variety as to how that picture is displayed. This transfor-
mation may be supported by the hardware which reads the display file and generates
the image. In the following chapters we shall see how other transformations may be
used in the picture-creation process.
We would like to give each segment its own image transformation attributes. Let
us now consider how this might be done. As we saw in the last chapter, our image
transformation can be specified by five numbers: x and y scale factors, a rotation
angle, and x and y translation amounts. These are, then, five more attributes to be
saved for each segment. In Chapter 4 we used global variables to hold the image trans-
formation parameters. Now, however, we shall use arrays so that individual parameters
may be stored for each of the display-file segments. We shall have an array for each
type of parameter as part of the segment table. Of course, the user must be able to set
the image transformation parameters. Translation may be set by the following al-
gorithm:

5.7 Algorithm SET-IMAGE-TRANSLATION(SEGMENT-NAME, TX, TY) User

routine to set the image translation for a segment
Arguments SEGMENT-NAME the segment being transformed
TX, TY the translation parameters
Global TRANSLATE-X, TRANSLATE-Y segment translation parameter table
Constant | NUMBER-OF-SEGMENTS the size of the segment table
BEGIN
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’:
TRANSLATE-X[SEGMENT-NAME] < TX;
TRANSLATE-Y [SEGMENT-NAME] <TY;
IF VISIBILITY[SEGMENT-NAME] THEN NEW-FRAME;
RETURN;
END;

The above algorithm saves translation amounts for the SEGMENT-NAME seg-
ment. Notice that a new-frame action is called only if the segment which is being mod-
ified happens to be visible.
A graphics system may not have individual routines for setting image scale and
rotation. This is because almost every time a scale or rotation is,performed, a transla-
tion adjustment is also needed. Our system will provide a single routine which sets all
of the image’s transformation parameters. (See Figure 5-5.)

5.8 Algorithm SET-IMAGE-TRANSFORMATION(SEGMENT-NAME, SX,

SY, A,
TX, TY) User routine to set the image transformation parameters of a segment
Arguments SEGMENT-NAME the segment being transformed
SX, SY, A, TX, TY the new image transformation parameters
Global VISIBILITY, SCALE-X, SCALE-Y, ANGLE, TRANSLATE-X,
TRANSLATE-Y arrays for attribute part of the segment table
Constant ©NUMBER-OF-SEGMENTS the size of the segment table
 SEGMENTS 157

FIGURE 5-5
Changing the image transformation.

BEGIN
IF SEGMENT-NAME < | OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
SCALE-X[SEGMENT-NAME] < SX;
SCALE-Y[SEGMENT-NAME] < SY;
- ANGLE[SEGMENT-NAME] < A;
TRANSLATE-X[SEGMENT-NAME] < TX;
TRANSLATE-Y [SEGMENT-NAME] < TY;
IF VISIBILIT Y[SEGMENT-NAME] THEN NEW-FRAME;
RETURN;
END;

REVISING PREVIOUS TRANSFORMATION

ROUTINES
We can modify the algorithms of Chapter 4 that set the image parameters to be compat-
ible with the segmented display file by making them correspond to the unnamed seg-
ment.

5.9 Algorithm TRANSLATE(TX, TY) (Upgrade of algorithm 4.6) Setting the transla-
tion parameters for the unnamed segment
Argument TX, TY the user translation specification
Global TRANSLATE-X, TRANSLATE-Y arrays for translation part
of the segment table
BEGIN
TRANSLATE-X[0]<— TX;
TRANSLATE-Y[0]< TY;
NEW-FRAME;
RETURN;
END;
158 CHAPTER FIVE

5.10 Algorithm SCALE(SX, SY) (Upgrade of algorithm 4.7) Image scaling

transformation
Arguments SX, SY the scaling parameters
Global SCALE-X, SCALE-Y the segment scaling parameter tables
BEGIN
SCALE-X[0]< SX;
SCALE-Y[0] < SY;
NEW-FRAME;
RETURN;
END;

5.11 Algorithm ROTATE(A) (Upgrade of algorithm 4.8) Image rotation

Argument Athe angle of rotation
Global ANGLE the segment-angle parameter table
BEGIN
ANGLE[0]< A;
NEW-FRAME;
RETURN;
END;

The algorithm for building the complete transformation matrix should now
specify which segment’s image transformation parameters should be used.

5.12 Algorithm BUILD-TRANSFORMATION (SEGMENT-NAME) (Upgrade of

algorithm 4.14) Build the image transformation matrix
Argument SEGMENT-NAME the segment which we are transforming
Global SCALE-X, SCALE-Y, ANGLE, TRANSLATE-X, TRANSLATE-Y
arrays for attribute part of the segment table
IMAGE-XFORM a 3 X 2 array containing the image transformation
INVERSE-IMAGE-XFORM a 3 X 2 array for the inverse of the image
transformation
BEGIN
IF SEGMENT-NAME < 0 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
IDENTIT Y-MATRIX(IMAGE-XFORM);
MULTIPLY-IN-SCALE(SCALE-X[SEGMENT-NAME], se -Y [SEGMENT-
NAME], IMAGE-XFORM);
MULTIPLY-IN-ROTATION(ANGLE[SEGMENT-NAME], IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(TRANSLATE-X[SEGMENT-NAME],
TRANSLATE-Y [SEGMENT-NAME], IMAGE-XFORM);
IDENTIT Y-MATRIX(INVERSE-IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(— WIDTH-START, — HEIGHT-START,
INVERSE-IMAGE-XFORM);
MULTIPLY-IN-SCALE(1 / WIDTH, | / HEIGHT, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION( — TRANSLATE-X[SEGMENT-NAME],
— TRANSLATE-Y[SEGMENT-NAME],INVERSE-IMAGE-XFORM);
MULTIPLY-IN-ROTATION( — ANGLE[SEGMENT-NAME], INVERSE-IMAGE-
XFORM);
 SEGMENTS 159

MULTIPLY-IN-SCALE(1/SCALE-X[SEGMENT-N AME],1/SCALE-Y[SEGMENT-
NAME], INVERSE-IMAGE-XFORM);
MULTIPLY-IN-SCALE(WIDTH, HEIGHT, INVERSE-IMAGE-XFORM);
MULTIPLY-IN-TRANSLATION(WIDTH-START,HEIGHT-START,
INVERSE-IMAGE-XFORM);
RETURN;
END;

We shall also need an initialization routine for this chapter. At the start of pro-
cessing, all segments should be empty. A call on DELETE-ALL-SEGMENTS ac-
complishes this. The unnamed segment should always be visible, and this attribute can
be initialized here. At the start of the program, no named segments should be open, so
the NOW-OPEN variable is initialized to 0.

5.13 Algorithm INITIALIZE-5

Global VISIBILITY the segment visibility table
NOW-OPEN the currently open segment
BEGIN
INITIALIZE-4;
DELETE-ALL-SEGMENTS;
VISIBILITY[0] <— TRUE;
NOW-OPEN < 0;
- RETURN;
END;

SAVING AND SHOWING SEGMENTS

So far we have givena lot of algorithms for creating and storing information about seg-
ments in a segment table. We still have to attach this segment structure to the routines
for saving and for showing display-file instructions. (See Figure 5-6.)
The first routine we must alter is PUT-POINT. We must add to this routine a
statement which increments the size of the segment currently open every time a new
instruction is added to a display file.

5.14 Algorithm PUT-POINT(OP, X, Y) Extension of algorithm 2.1 to include updating

the segment table
Arguments OP, X, Y a display-file instruction
Global NOW-OPEN the segment currently open
SEGMENT-SIZE the segment size array
DF-OP, DF-X, DF-Y the three display-file arrays
FREE the position of the next free display-file cell
BEGIN
SEGMENT-SIZE[NOW-OPEN] < SEGMENT-SIZE[NOW-OPEN] + 1;
IF FREE > DFSIZE THEN RETURN ERROR ‘DISPLAY FILE FULL;
DF-OP[FREE] < OP;
DF-X[FREE] < X;
DF-Y[FREE] < Y;
160 = cHAPTER FIVE

Segment
table

User program

FIGURE 5-6
Picture generation with segments.

FREE < FREE + 1;

RETURN;
END;

The second routine we must change is MAKE-PICTURE-CURRENT. This

routine contained an instruction to display the entire display file, beginning with in-
struction 1 and ending with index FREE — 1. We replace this statement by a loop
which steps through the segment table, examining each segment to see if its size is
greater than O and if it is also visible. If both of these conditions are met, then its
image transformation matrix is formed and the segment is interpreted. The segment
table tells us where to begin and how many instructions to interpret. We pass this infor-
mation to our INTERPRET routine as its arguments.(See Figure 5-7.)

5.15 Algorithm MAKE-PICTURE-CURRENT (Revision ofalgorithm 4.10)

Global SEGMENT-START, SEGMENT-SIZE, VISIBILITY the segment table
ERASE-FLAG a flag indicating that the display should be erased
Local I a variable for stepping through the segment table
Constant . NUMBER-OF-SEGMENTS the size of the segment table
BEGIN
IF ERASE-FLAG THEN
BEGIN
ERASE;
ERASE-FLAG < FALSE;
END;
 SEGMENTS 161

FOR I = 0 TO NUMBER-OF-SEGMENTS DO
IF SEGMENT-SIZE[I] + 0 AND VISIBILITY[I] THEN
BEGIN
BUILD-TRANSFORMATION(I);
INTERPRET(SEGMENT-START[I], SEGMENT-SIZE[I]);
END;
DISPLAY;
DELETE-SEGMENT(0);
RETURN;
END;

OTHER DISPLAY-FILE STRUCTURES

The necessary operations on the display file are insertion, when we construct a draw-
ing; selection, when we interpret and display; and deletion, when we are finished with
a segment. There are many possible data structures which might be used. We have cho-
sen a simple one, the array. While insertion, selection, and deletion are easy for an
array, deletion may not be very efficient. If we wish to remove an instruction at the be-
ginning of the display file, we must move all succeeding instructions. If the display
file is large, this could mean a lot of processing to recover only a small amount of
storage. One alternative data structure which might be used is the linked list. (See
Figure 5-8.)
In_a linked list the instructions are not stored in order; rather a new field is added
to the instruction. This field, called the link or pointer, gives the location of the next in-
struction. We step through the instructions by following the chain of links. The instruc-
tion cells which have not yet been used are also linked to form alist of available space.

Segment table

rg

/ he
/|
: %/
Pao a4 hs

FIGURE 5-7
The segment table indicates the portions of
the display file used to construct the picture.
162 CHAPTER FIVE

wmoON
CoO
fF
Dn
WN

oO

Start
FIGURE 5-8 /
The display file as a linked list (drawn three different ways).

When a new instruction is added to a display file, a cell is obtained from the list of
available space, the correct instruction operation code and operands are stored, and the
cell is linked to the display-file list. Deletion of cells from a linked list is very easy. To
remove a cell, we need only change the pointer which points to that cell so that it
points to the succeeding cell. (See Figure 5-9.)
We could also link the cell which we have removed to the list of available space.
In the linked list scheme, deleting the cell means changing two links. This can be
much more efficient than moving a number of instructions, as we did for the array. The
disadvantages of the linked list scheme are that it requires more storage to hold the
links and that it is costly to locate arbitrary cells.
A third scheme, which is between the array and linked list methods, is a paging
scheme. In this method the display file is organized into a number of small arrays
called pages. The pages are linked to form alinked list of pages. Each segment begins
at the beginning of a page. If a segment ends at some point other than a page boundary,

FIGURE 5-9
Deleting display-file instructions from a linked list.
 SEGMENTS 163

then the remainder of that page is not used. In this scheme, display-file instructions
can be accessed within a page just as they were accessed in an array. When the end of a
page is reached, a link is followed to find the next page. (See Figure 5-10.)
By grouping the instructions into pages, we have reduced the number of links in-
volved, yet we are still able to delete a segment by simply altering links. A list of un-
used or available pages provides a source of new pages when the display file is ex-
tended, and deleted pages may be returned to this list for reuse. Some disadvantages of
this scheme are that storage is lost at the end of the page if a segment does not com-
pletely fill it and that accessing is a bit more complex.
In the above discussion, once a display-file segment has been closed, it can no
longer be altered. There is no way to replace display-file instructions, and once closed,
the segment cannot even be extended. While we have built this restriction into our sys-
tem, other systems might allow such operations (although some questions may arise,
for instance, whether modification commands should follow the current attributes,
such as line style and character spacing, or those that were in effect in the original seg-
ment at the point of modification). When editing of the display file is allowed, a linked
structure may be much more natural than our array scheme. The extension of our array
scheme is considered in Programming Problem 5-12.
There are many other possible storage schemes besides the three we have men-
tioned. We have tried to isolate access to the display file in the routines GET-POINT
and PUT-POINT, although the segment deletion routine also depends on the display-
file structure. We have organized our program in this way so that different display-file
data structures might be employed with a minimum of alteration to the algorithms.

SOME RASTER TECHNIQUES

In the graphics package we are developing, the only method available for altering an
image is to change the display file or image transformation and reinterpret the entire
picture. This method works well for storage tube and vector refresh displays, but may
prove inefficient for raster displays, where clearing and recomputing the pixel values
for the entire frame buffer can be costly. The principle behind fast modification of a
raster display is to change as little as possible. Special techniques have been developed
for raster displays which allow altering a portion of the display while leaving the re-

L
FIGURE 5-10
Linked pages of display-file instructions.
164 CHAPTER FIVE

mainder of the frame buffer intact. For example, suppose we wished to make one seg-
ment invisible while maintaining the current status on all remaining segments. Instead
of clearing the frame buffer and redrawing the entire picture, we might just redraw the
segment which we wish to make invisible, only in this drawing use a background value
for the setting of each pixel. This in effect erases all lines which were previously drawn
by the segment. The segment would be removed without clearing the entire display.
This technique could leave gaps in lines belonging to other segments if the segments
shared pixels with the invisible segment (for example, points where a line from the in-
visible segment crosses a line from the visible segment). This damage could be re-
paired by reinterpreting the segments which are still visible. Again this can be done
without first clearing the frame buffer. (See Figure 5-11.)
Another frame buffer technique can sometimes be used for efficient translation
of an image. The image may be moved from one portion of the screen to a different
portion of the screen by simply copying those pixels involved in the image from one
position in the frame buffer to their new position. If the image is confined to a box,
then only those pixels within that box need be copied. Pixels outside can be left un-
changed. This could be much more efficient than setting all pixels in the frame buffer
to their background value and recomputing the pixel settings for the translated image.
(See Figure 5-12.)
There is an operation called a RasterOp or bit block-transfer (BITBLT) which can
be quite useful in working with raster displays. The idea is to perform logic operations
on bit-addressed blocks of memory. The BITBLT works on subarrays of the frame buf-
fer. It performs simple operations on subarrays (such as turning all pixels on or
off, shifting all values by a row or a column, and copying values from another subar-
ray); and for one-bit-per-pixel frame buffers, it forms the logical AND, OR, or
EXCLUSIVE-OR of pixel values in two subarrays. For example, we might use a
BITBLT to copy a rectangular area of the display to some other area on the screen. Or
we might use the BITBLT to clear a portion of the display. We have already introduced

Before

FIGURE 5-11
After Removing an image by redrawing with the background.
 SEGMENTS 165

FIGURE 5-12
Alter only the pixels contained by the box.

this idea in connection with character generation. Pixel values for a character may be
copied into a subarray of the frame buffer from some fixed template. BITBLT
operations may be implemented in hardware so that they are very fast. (See Figure
5-13.)

AN APPLICATION
Let us suppose that a computer graphics display system is to be used to aid in the dock-
ing of a large ship. The relative position of the ship and the dock (measured by the
ship’s sensors) is to be presented graphically by showing an image of the ship and dock
as they might be seen from above. How might such a program be written in our sys-
tem? We can use our LINE and POLYGON commands to generate images for the dock
and for the ship. If the instructions for these images are placed in different display-file

FIGURE 5-13
A ANDB Logical operations on scenes.
166 = CHAPTER FIVE

Ship

FIGURE 5-14
oe Graphical display of ship’s position during docking.

segments, then we can use the image transformation to position each independently.
(See Figure 5-14.)

CREATE-SEGMENT(1);
draw the ship
CLOSE-SEGMENT;
CREATE-SEGMENT(2);
draw the dock
CLOSE-SEGMENT;

Now we need a loop to repeatedly update the ship’s position on the display. This
loop will obtain the position ’from the ship’s sensors and determine the rotation and
translation parameters required. These parameters are used in the image transformation
to place the ship and dock in their correct positions on the display. Of course, if the
ship’s position has not altered since the last check, then no updating is necessary. We
can extend this program to handle docking at several ports-of-call by entering, in sepa-
rate segments, the shape of the dock at each port. We then make the dock at the current
port-of-call visible, while all the others are made invisible.

FURTHER READING
A segmented display file and the operations on it are described in [NEW74]. A system
which builds pictures from subpictures composed of instances of segments is described
in [JOS84]. An overview of several segmentation schemes is given in [FOL76]. Spe-
cial hardware which supports BITBLT or RasterOp functions on 8 X 8 pixel squares is
described in [SPR83]. Techniques for producing animation on different hardware ar-
chitectures are described in [BAE79]. A bibliography on computer animation is given
in [THA8S5].

[BAE79] Baecker, ‘‘Digital Video Display Systems and Dynamic Graphics,’’ Computer Graphics, vol. 13,
no. 2, pp. 48-56 (1979).
[FOL76] Foley, J. D., ‘‘Picture Naming and Modification: An Overview,’’ Computer Graphics, vol. 10, no.
1, pp. 49-53 (1976).
 SEGMENTS 167

[JOS84] Joshi, R. R., Arunachalam, H., ‘‘Multilevel Picture Segmentations in the Graphics System—
GRASP,”’ Computers & Graphics, vol. 8, no. 2, pp. 167-176 (1984).
[NEW74] Newman, W. M., and Sproull, R. FE, ‘‘An Approach to Graphics System Design,’’ Proceedings of
the IEEE, vol. 62, no. 4, pp. 471-483 (1974).
[SPR83] Sproull, R. E, Sutherland, I. E., Thompson, A., Gupta, S., Minter, C., ‘‘The 8 by 8 Display,”’
ACM Transactions on Graphics, vol. 2, no. 1, pp. 32-56 (1983).
[THA85] Thalmann, N., Thalmann, D., ‘‘An Indexed Bibliography on Computer Animation,’’ JEEE Com-
puter Graphics and Applications, vol. 5, no. 7, pp. 76-86 (1985).

EXERCISES
5-1 Suppose that the segment table contains room for only five named segments and the un-
named segment. Describe the segment table built by each of the following program fragments.
a) INITIALIZE-5
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
LINE-ABS-2(0.0, 1.0)
LINE-ABS-2(0.5, 0.5)
b S INITIALIZE-5
CREATE-SEGMENT(1)
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
LINE-ABS-2(0.0, 1.0)
y LINE-ABS-2(0.5, 0.5)
CLOSE-SEGMENT
c) INITIALIZE-5
CREATE-SEGMENT(4)
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
LINE-ABS-2(0.0, 1.0)
LINE-ABS-2(0.5, 0.5)
CLOSE-SEGMENT
d wa INITIALIZE-5
CREATE-SEGMENT(3)
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
CLOSE-SEGMENT
LINE-ABS-2(0.0, 1.0)
LINE-ABS-2(0.5, 0.5)
e ~ INITIALIZE-5
MOVE-ABS-2(0.5, 0.5)
CREATE-SEGMENT(2)
LINE-ABS-2(0.5, 1.0)
LINE-ABS-2(0.0, 1.0)
CLOSE-SEGMENT
LINE-ABS-2(0.5, 0.5)
f) INITIALIZE-5 _
CREATE-SEGMENT(2)
168 CHAPTER FIVE

MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
CLOSE-SEGMENT
CREATE-SEGMENT(4)
LINE-ABS-2(0.0, 1.0)
LINE-ABS-2(0.5, 0.5)
CLOSE-SEGMENT
8 INITIALIZE-5
wa

CREATE-SEGMENT(2)
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
CLOSE-SEGMENT
CREATE-SEGMENT(4)
LINE-ABS-2(0.0, 1.0)
LINE-ABS-2(0.5, 0.5)
CLOSE-SEGMENT
DELETE-SEGMENT(4)
h a INITIALIZE-5
CREATE-SEGMENT(2)
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
CLOSE-SEGMENT
CREATE-SEGMENT(4)
LINE-ABS-2(0.0, 1.0)
LINE-ABS-2(0.5, 0.5) /
CLOSE-SEGMENT
DELETE-SEGMENT(2)
i) INITIALIZE-5
CREATE-SEGMENT(2)
MOVE-ABS-2(0.5, 0.5)
LINE-ABS-2(0.5, 1.0)
CLOSE-SEGMENT
RENAME-SEGMENT(2, 4)
5-2 Suggest other attributes besides visibility and the image transformation which might be as-
sociated with a display-file segment.
5-3 In our implementation, we have used one segment-table entry for each possible name, so
that if we want 1000 possible names, arrays of size 1000 would be needed. Suppose, however,
that we wish to allow the user to use any name between | and 10,000, but restrict him to using
no more than 100 segments at any one time. Suggest a segment-table organization which will
allow him to do this using arrays of only 100 elements.
5-4 Suggest other possible data structures which could be used for holding display-file seg-
ments. Discuss their advantages and disadvantages.
5-5 Outline PUT-POINT, GET-POINT, and DELETE-SEGMENT algorithms for a linked list
display-file data structure.
5-6 Consider a graphics system which allows the user to have more than one segment open at a
time. Suggest how we might implement such a system. What would be the advantages and dis-
advantages?
 SEGMENTS 169

5-7 We have mentioned the binary operations of AND, OR, and EXCLUSIVE-OR as being use-
ful in combining raster images. Generalize these operations to gray-level displays, where each
pixel has a multibit intensity value.
5-8 List all of the possible logical operations which can be used to combine two binary raster
images and describe their effects. (The full list contains 16 operations.)

PROGRAMMING PROBLEMS
5-1 Implement algorithms 5.1 through 5.15 to extend the system to allow segmentation of the
display file.
5-2 Write the routines
INQUIRE-SEGMENT-OPEN(SEGMENT-NAME)
INQUIRE-VISIBILITY(SEGMENT-NAME, ON-OFF)
INQUIRE-IMAGE-TRANSFORMATION(SEGMENT-NAME, SX, SY, A, TX, TY)
which return to the user the name of the currently open segment, the visibility status of a seg-
ment, and the image transformation parameters for a segment.
§-3 Write a subroutine DIE(N) which displays the image of a die showing N dots (N ranges be-
tween | and 6). Use one segment to draw the cube

and four other segments with the dot patterns

fi dq CT 7 a ah ip “a

L a cE J e et L aa

Display N dots by setting the visibility attributes of the four dot-pattern segments. For example,
if N were 5, the first three patterns should be visible and the fourth invisible, giving

which when drawn on the cube background gives

170 CHAPTER FIVE

5-4 Consider the map of Programming Problem 3-7. We wish to draw this map again (unfilled
this time) but with other information superimposed. We would like to show the map as follows:
a) With the location of widget distributors
b) With the locations of the competitive doohickey distributors
c) With the locations of both
To do this, put the map in one display-file segment, markers for widget locations in a second
segment, and different markers for doohickey locations in a third segment. Produce the approp-
riate maps by setting the segment’s visibility attribute. The distributors should be shown at the
following locations.
Widget (0.2, 0.7), (0.3, 0.5), (0.8, 0.8), (0.6, 0.5)
Doohickey (0.8, 0.2), (0.2, 0.3), (0.3, 0.7), (0.7, 0.7)
5-5 Write a subprogram which will draw four images to give a kaleidoscope effect. Each image
will be formed by redrawing the picture with a new display image transformation. The first
image will have the identity transformation, the second will be reflected in the line x = 0.5, the
third will be reflected in the line y = 0.5, and the fourth will be reflected in the point (0.5, 0.5).
5-6 Consider the polygon command for drawing a square with vertices (0.0), (0.1), (1.1), (1.0),
and an image transformation which will shrink the square by 0.8, rotate it by 7/5, and center it
at (0.5, 0.5). For each segment you have available, construct a square which is smaller and
further rotated than that in the previous segment. Show all segments.
5-7 Write a subroutine PIE-CHART(FA, N, R) which takes as arguments an array FA of frac-
tions, a number N which tells how many fractions are in this array, and a radius R. The program
should draw separate pie-shaped wedges for each fraction which fit together in a circle. Con-
struct each wedge byacall to the PIE routine of Programming Problem 4-5. Each wedge should
be in its own display-file segment so that it can have its own image transformation. As an exam-
ple, suppose that FA contained 0.3, 0.5, and 0.2, then PIE-CHART(FA, 3, 0.4) should look like
this:

5-8 Construct a graphing program that allows the user to view several graphs either indepen-
dently or overlaid.
5-9 a) Construct a face using separate segments for the head, eyes, nose, and mouth.
b) Change the face by applying image transformations to the features.
5-10 Construct a cartoon character which can be shown in different positions on different back-
grounds (using the visibility and image transformation attributes of segments). Allow the charac-
ter to change by placing parts of the character in different segments.
5-11 Show an ‘‘exploded’’ view of the components which make up some object. Place
each
component in a separate segment and allow the user to enlarge it.
 sEGMENTS 171

*5-12 There are two routines which are very useful for some applications. The first routine
COPY-SEGMENT(SEGMENT-1, SEGMENT-2) makes SEGMENT-2 draw the same picture
(possibly with a different image transformation) as SEGMENT-1. One way to do this is to actu-
ally copy all the instructions in SEGMENT-1 to the end of the display file and associate these
duplicate instructions with SEGMENT-2. Another method requiring less storage is to give
SEGMENT-2 the same display-file starting index as SEGMENT-1. In this case, the instructions
are entered only once, but both segments use them. The problem with this approach is that the
DELETE-SEGMENT routine becomes a little more complex. A second useful routine is
EXTEND-SEGMENT(SEGMENT). This routine reopens a segment which has been closed. It
does not delete any of the existing instructions in that segment, but allows new instructions to be
added.
a) Design and implement a COPY-SEGMENT routine.
b) Design and implement an EXTEND-SEGMENT routine.
*5-13 Consider the simple animation problem of a square wheel rolling across the screen,
obscuring a row of houses in the background as it rolls.
a) Determine the timings of the basic operations on your system (arithmetic operations,
procedure calls, frame buffer rendering, etc.). Using these timings, analyze the algorithms in-
volved and estimate how many frames per second your system can produce (30 frames per sec-
ond are required for smooth animation).
b) Implement the animation problem and compare its speed with the result of your calcu-
lation.
*5-14 It is often useful to be able to save an image. A drawing could be saved and later
examined. If the image is saved in the form of a display file, then when redisplayed, the visibil-
ity and image transformation attributes can be used to provide desired versions. Retrieved im-
ages can also be modified by adding or deleting segments. Extend your system to allow the sav-
ing of a display-file image. The system should be capable of reloading and displaying the stored
image. The user should be able to load from a library of saved images.
WINDOWING
AND
CLIPPING

INTRODUCTION
An architect may have a graphics program to draw an entire building but be interested
in only the ground floor. A businessman may have a map of sales for the entire nation
but be interested in only the northeast and southwest. An integrated circuit designer
may have a program for displaying an entire chip but be interested in only a few regis-
ters. Often, the computer is used in design applications because it can easily and accu-
rately create, store, and modify very complex drawings. When drawings are too com-
plex, however, they become difficult to read. In such situations it is useful to display
only those portions of the drawing that are of immediate interest. This gives the effect
of looking at the image through a window. Furthermore, it is desirable to enlarge these
portions to take full advantage of the available display surface. The method for select-
ing and enlarging portions of a drawing is called windowing. The technique for not
showing that part of the drawing which one is not interested in is called clipping. (The
term windowing has been used to mean several different things, all related to partition-
ing of the display. A single consistent definition is used within this text, but other defi-
nitions are found in the literature and may be a source of confusion.)
In this chapter we shall consider the ideas of windowing, clipping, and viewing
transformation. We will add to our graphics package the routines for setting windows
and viewports and for removing lines which lie outside the region we wish to display.
The transformation routines of Chapter 4 will guide us in setting up a viewing transfor-
mation which will be applied to each display-file instruction as the instruction is
created.

172
 WINDOWING AND CLIPPING 173

THE VIEWING TRANSFORMATION

It is often useful to think of two models of the item we are displaying. There is the ob-
ject model and there is the image of the object which appears on the display. When we
speak of the object, we are actually referring to a model of the object stored in the com-
puter. The object model is said to reside in object space. This model represents the ob-
ject using the physical units of length. In the object space, lengths of the object may be
measured in any units from light-years to angstroms. The lengths of the image on the
screen, however, must be measured in screen coordinates (we have normalized the
screen coordinates so that they range between 0 and 1). (See Figure 6-1.)
We must have some way of converting from the object space units of measure to
those of the image space (screen space). This can be done by the scaling transformation
of Chapter 4. By scaling, we can uniformly reduce the size of the object until its di-
mensions lie between 0 and 1. Very small objects can be enlarged until their overall di-
mension is almost | unit. The physical dimensions of the object are scaled until they
are suitable for display. (See Figure 6-2.)
It may be, however, that the object is too complex to show in its entirety or that
we are particularly interested in just a portion of it. We would like to imagine a box
about a portion of the object. We would only display what is enclosed in the box. Such
a box is called a window. (See Figure 6-3.)
It might also happen that we do not wish to use the entire screen for display. We
would like to imagine a box on the screen and have the image confined to that box.
Such a box in the screen space is called a viewport. (See Figure 6-4.)
When the window is changed, we see a different part of the object shown at the
same position on the display. (See Figure 6-5.) If we change the viewport, we see the
same part of the object drawn at a different place on the display. (See Figure 6-6.)

12M

—=——12M——>}
Object space

FIGURE 6-1
ie ia
G2S;.2) (Go; <2) In the object space, position is measured in physical units, such as
meters. In the image space, position is given in normalized screen
Image space coordinates.
174s cuaprer SIX

1/48 0 O

x |.0 1/48 0
OP =LOr eet

FIGURE 6-2
A scaling transformation will convert object coordinate
units to normalized screen coordinates.

In specifying both window and viewport, we have enough information to deter-

mine the translation and scaling transformations necessary to map from the object
space to the image space. This can be done with the following three steps. First, the
object together with its window is translated until the lower-left corner of the window
is at the origin. Second, the object and window are scaled until the window has the di-
mensions of the viewport. In effect, this converts object and window into image and

| Window
|

SS eee ace
|
Object space

FIGURE 6-3
Image space A window to view only part of an object.
 WINDOWING AND CLIPPING 175

Viewport

FIGURE 6-4
A viewport to define the part of the screen to be used.

viewport. The final transformation step is another translation to move the viewport to
its correct position on the screen. (See Figure 6-7.)
We are really trying to do two things. We are changing the window size to be-
come the size of the viewport (scaling) and we are positioning it at the desired location
on the screen (translating). The positioning is just moving the lower-left corner of the
window to the viewport’s lower-left corner location, but we do this in two steps. We
first move the corner to the origin and second move it to the viewport corner location.
We take two steps because while it is at the origin, we can perform the necessary scal-
ing without disturbing the corner’s position.
The overall transformation which performs these three steps we shall call the
viewing transformation. It creates a particular view of the object. (See Figure 6-8.)
Let us consider an example of a viewing transformation. If our window has left
and right boundaries of 3 and 5 and lower and upper boundaries of 0 and 4, then the
first translation matrix would be
’ 100
G-1 0
4-011
Suppose the viewport is the upper-right quadrant of the screen with boundaries at
0.5 and 1.0 for both x and y directions. The length of the window is 5 — 3 = 2 in the

FIGURE 6-5
Different windows, same viewports.
 FIGURE 6-6
Same windows, different view-
ports.

Translate

Translate

FIGURE 6-7
Steps in the viewing transformation.

176
 WINDOWING AND CLIPPING 177

FIGURE 6-8
The viewing transformation.

x direction. The length of the viewport is 1.0 — 0.5 = 0.5, so the x scale factor is
0.5/2 = 0.25. In the y direction, we have the factor 0.5/4 = 0.125, so the scaling
transformation matrix will be

022520 0
Di nocOA25°0 +
Bile iG) 1 ERY AREINE Dee
|
as bsbeta wD)
_ Finally, to position the viewport requires a translation of
fos / i
ape
05 0.5 1 Ag
3 |
Parca(5aX ||
The viewing transformation is then i WW 3

70.0 un 0 1 ee 0.25 0 0
er OO ee les Orla Or ex dees Oli Onin 0-125 iOloO-D
paren 1 0 0 1 0:5..cOSeuk =0.25 0.5 1

In general, the viewing transformation is

1 0 0| | (VXH-VXL) 0 0 1 Gn 20
(WXH- WXL)
(VYH ~VYL 0 D5 20
: fo (WYH- WYL)
eee CO

-WYL 1 0 0 1 Vole VY)

-WXL

(WXH- WXL)
(VYH- VYL) 0 (6.2)
. 0 (WYH- WYL)
(VXH-VXL) yyy, wy {YYH=VYL)
VXL" WXL (xXH-WXL) (WYH- WYL)
178 CHAPTER SIX

The variables have been named according to the rule that V stands for viewport and W
for window, X for the position of a vertical boundary and Y for a horizontal boundary,
H for the high boundary and L for the low boundary. .
Note that if the height and width of the window do not have the same proportions
as the height and width of the viewport, then the viewing transformation will cause
some distortion of the image in order to squeeze the shape selected by the window into
the shape presented by the viewport.

VIEWING TRANSFORMATION
IMPLEMENTATION
The first step of our viewing transformation is specifying the size of the window. We
will confine our window to a rectangular shape parallel with the x and y axes. By
doing this, we need only specify the smallest and largest possible x values and the
smallest and largest possible y values. Our routine for specifying the size of the win-
dow will store these boundary values in global variables so that they will be available
when it comes time to perform the transformation. Likewise, we must specify the
boundaries of the viewport. The length of the viewport or window in either the x or the
y dimension is determined by subtracting the lower boundary from the upper boundary.
Note that we cannot have the lower boundary equal to the upper boundary for a win-
dow because this would cause us to divide by zero when we try to determine the scal-
ing transformation. Algorithms for setting the viewport and window dimensions are
given below.

6.1 Algorithm SET-VIEWPORT(XL, XH, YL, YH) User routine for specifying the
viewport
Arguments XL, XH the left and right viewport boundaries
YL, YH the bottom and top viewport boundaries
Global VXL-HOLD, VXH-HOLD, VYL-HOLD, VYH-HOLD
storage for the viewport boundaries
BEGIN
IF XL = XH OR YL = YH THEN RETURN ERROR ‘BAD VIEWPORT’;
VXL-HOLD < XL;
VXH-HOLD < XH;
VYL-HOLD < YL;
VYH-HOLD < YH;
RETURN;
END;

6.2 Algorithm SET-WINDOW(XL, XH, YL, YH) User routine for specifying the
window
Arguments XL, XH the left and right window boundaries
YL, YH the bottom and top window boundaries
Global WXL-HOLD, WXH-HOLD, WYL-HOLD, WYH-HOLD
storage for the window boundaries
BEGIN
IF XL = XH OR YL = YH THEN RETURN ERROR ‘BAD WINDOW’;
WXL-HOLD < XL;
 WINDOWING AND CLIPPING 179

WXH-HOLD < XH;

WYL-HOLD < YL; Je Fun — §/
WYH-HOLD < YH; ;
RETURN;
END;

In our system we shall not change viewing parameters in the middle of a\seg-.)
ment. Each segment is treated as a snapshot of the object. The viewing transformation
describes how the camera is positioned. In this model it is reasonable to prohibit move-
ment of the camera while taking the picture. We shall follow this rule by keeping two
sets of viewing parameters, one set for the user to change and a second set to actually
be used in the windowing and clipping routines. Changing the window in our system
becomes a two-step process. First, the user changes his set of window and viewport
boundaries. Second, the user’s values are copied into the variables actually used by the
windowing and clipping routines. By performing this copying as part of the segment-
creation process, we ensure that changes in the viewing parameters being used cannot
occur in the middle of a segment. (See Figure 6-9.)
We should note that this restriction on changing viewing parameters is just a rule
for the particular viewing model we have chosen. There is no fundamental reason why
a system could not be written which would allow changing the view at any point; in
fact, while the CORE graphic system matches our approach, the GKS system does
allow change of the viewing transformation within segments.
“We shall write an algorithm for copying the user’s specifications into the sys-
tem’s viewing parameters. This routine also calculates the window-to-viewport scale
factors.

6.3 Algorithm NEW-VIEW-2 Set the clipping and viewing parameters from the current
window and viewport specifications
Global WXL-HOLD, WYL-HOLD, WXH-HOLD, WYH-HOLD
the user’s window parameters
VXL-HOLD, VYL-HOLD, VXH-HOLD, VYH-HOLD
the user’s viewport parameters
a WXL, WYL, WXH, WYH, VXL, VYL, VXH, VYH
the current clipping parameters
WSX, WSY the window-to-viewport scale factors

Parameters Parameters
specified used in
by user viewing

VXL-HOLD WXL-HOLD
SET-VIEWPOINT VXH-HOLD WXH-HOLD VIEWING-
SET-WINDOW VYL-HOLD WYL-HOLD TRANSFORMATION
VYH-HOLD WYH-HOLD

NEW-VIEW-2
ae ae.
FIGURE 6-9
viewing.
A NEW-VIEW action is required to make the user’s specification that which is used in
180 = cHAPTER SIX

BEGIN
WXL <— WXL-HOLD;
WYL <— WYL-HOLD;
WXH < WXH-HOLD;
WYH <—WYH-HOLD;
VXL <— VXL-HOLD;
VYL <— VYL-HOLD;
VXH < VXH-HOLD;
VYH < VYH-HOLD;
WSX <— (VXH — VXL)/(WXH — WXL);
WSY <—(VYH — VYL)/(WYH — WYL);
RETURN;
END;

We wish any given window setting to apply to an entire display-file segment. We

can enforce this restriction by only allowing the above copying of parameters to occur
when a segment is created. We shall therefore modify our segment-creation routine to
reset the viewing transformation to match the latest user request.

6.4 Algorithm CREATE-SEGMENT(SEGMENT-NAME) (Modification of

Algorithm 5.1) User routine to create a named segment
Argument SEGMENT-NAME the segment name
Global NOW-OPEN the segment currently open
FREE the index of the next free display-file cell
SEGMENT-START, SEGMENT-SIZE, VISIBILITY ANGLE, SCALE-X,
SCALE-Y, TRANSLATE-X, TRANSLATE-Y the segment-table arrays
Constant .NUMBER-OF-SEGMENTS size of the segment table
BEGIN
IF NOW-OPEN > 0 THEN RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAMB’;
IF SEGMENT-SIZE[SEGMENT-NAME] > 0 THEN
RETURN ERROR ‘SEGMENT ALREADY EXISTS’;
NEW-VIEW-2;
SEGMENT-START[SEGMENT-NAME] < FREE;
SEGMENT-SIZE[SEGMENT-NAME] < 0;
VISIBILITY [SEGMENT-NAME] < VISIBILITY [0];
ANGLE[SEGMENT-NAME] <— ANGLE[0];
SCALE-X[SEGMENT-NAME] <— SCALE-X[0];
SCALE-Y[SEGMENT-NAME] < SCALE-Y[0];
TRANSLATE-X[SEGMENT-NAME] < TRANSLATE-X[0];
TRANSLATE-Y[SEGMENT-NAME] <— TRANSLATE-Y[0];
NOW-OPEN <- SEGMENT-NAME;
RETURN;
END;

We wish to perform the following transformations. First, we wish to translate by

the lower x and y boundaries of the window. This moves the lower-left corner of the
window to the origin. Second, we wish to scale by the size of the viewport divided by
 WINDOWING AND CLIPPING 18]

the size of the window. This changes the dimensions of the window to those of the
viewport. Finally, we wish to translate by the lower x and y boundary values of the
viewport. This moves the lower-left corner from the origin to the correct viewport posi-
tion. We can form each of these transformation matrices as we did in Chapter 4. We
can multiply the matrices together to form a single transformation (Equation 6.2) and
then apply it to a general point. This yields the following viewing transformation al-
gorithm:

6.5 Algorithm VIEWING-TRANSFORM(OP, X, Y) Viewing transformation of a

point
Arguments OP, X,Y the instruction to be transformed
Global WXL, WYL, WSX, WSY, VXL, VYL window and viewport parameters
Local X1, Y1 the transformed point
BEGIN
X1 <— (X — WXL) * WSX + VXL;
Y1 <—(Y — WYL) * WSY + VYL;
PUT-POINT(OP, X1, Y1);
RETURN;
END;

Note that the above algorithm not only performs the viewing transformation but
also enters the resulting instruction into the display file. The display file will hold the
image space model.

CLIPPING
Now that we have seen how our picture may be correctly scaled and positioned, we
shall consider how to cut off the lines which are outside the window so that only the
lines within the window are displayed. This process is called clipping. In clipping we
examine each line of the display to determine whether or not it is completely inside the
window, lies completely outside the window, or crosses a window boundary. If it is in-
side, the line is displayed; if it is outside, nothing is drawn. If it crosses the boundary,
we must determine the point of intersection and draw only the portion which lies in-
side. (See Figure 6-10.)

ex
et pies
Before After

FIGURE 6-10
Clipping.
182 = cHapTER SIX

Different graphic elements may require different clipping techniques. A charac-

ter, for example, may be either entirely included or omitted depending on whether or
not its center lies within the window. This technique will not,work for lines, and some
methods used for lines will not work for polygons.

THE COHEN-SUTHERLAND OUTCODE

ALGORITHM
A popular method for clipping lines is the Cohen-Sutherland Outcode algorithm. The al-
gorithm quickly removes lines which lie entirely to one side of the clipping region
(both endpoints above, or below, or right, or left). The algorithm makes clever use of
bit operations (outcodes) to perform this test efficiently. Segment endpoints are each
given 4-bit binary codes. The high-order bit is set to 1 if the point is above the window;
the next bit is set to 1 if the point is below the window; the third and fourth bits indi-
cate right and left of the window, respectively. The lines which form the window boun-
dary divide the plane into nine regions with the outcodes shown in Figure 6-11.
If the line is entirely within the window, then both endpoints will have outcode
0000. Segments with this property are accepted (segment ST in Figure 6-12). If the
line segment lies entirely on one side of the window (say entirely above it), then both
endpoints will have a | in the outcode bit position for that side (the first bit will be 1
for both endpoints). We can check to see if the line is entirely on one side of the win-
dow by taking the logical AND of the outcodes for the two endpoints. If the result of
the AND operation is nonzero, then the line segment may be rejected. Thus one test de-
cides if the line segment is entirely above, or entirely below, or entirely to the right, or
entirely to the left of the window. For example, segments AB and CD in Figure 6-12
would be quickly removed.
The difficult cases occur when a line crosses one or more of the lines which con-
tain the clipping boundary (such as segments EF and IJ). For these cases, the point of

FIGURE 6-11
Outcodes for the plane.
 WINDOWING AND CLIPPING 183

FIGURE 6-12
Testing and dividing line segments.

intersection between the line segment and clipping boundary lines may be used to
break up the line segment. The resulting pieces may be tested for acceptance or rejec-
tion. Segment EF may be broken into EG and GF, where EG lies above and GF lies to
the right, so both would be rejected. Segment IJ might be divided into IK and KJ. IK
can be rejected because it lies to the left, but KJ must be further divided. Forming KL
and LJ, we see that KL may be rejected as lying above but LJ must still be divided into
LM and MJ. LM is contained and accepted, while MJ is to the right and rejected. At
worst, the intersections with all four boundary lines will be calculated in order to clip
the line.
The following is a brief outline of the algorithm (the details are left as an exer-
cise): First, we compute the outcodes for the two endpoints (p, and p.) of the segment.
Next, we enter a loop. Within the loop we check to see if both outcodes are zero; if so,
we enter the segment into the display file, exit the loop, and return. If the outcodes are
not both zero, then we perform the logical AND function and check for a nonzero re-
sult. If this test is nonzero, then wereject the line, exit the loop, and return? If neither
of these tests is satisfied, we must subdivide the line segment
and repeat the loop. If
the outcode for p, is zero, exchange the points p, and p, and also their outcodes. Find
a nonzero bit in the outcode of pj. If it is the high-order bit, then find the intersection
of the line with the top boundary of the window. If it is the next bit position, then sub-
divide along the bottom boundary. The other two bits indicate that the right and left
boundaries should be used. Replace the point p, with the intersection point and calcu-
late its outcode. Repeat the loop.
fh ‘ =e t , {
ry

bh QA }Va 5 hy C/YNS I> 8

pf yon SMV

CON NMr r / Ww S Sen o Cohky~

184 cHapTER SIX

THE SUTHERLAND-HODGMAN ALGORITHM

The Cohen-Sutherland algorithm works well for lines, but we would like a method
which may be used with polygons:as well. Our clipping routines will be based on a
method discovered by Sutherland and Hodgman. The method unbundles the clipping
test to clip against each of the four boundaries individually. The idea behind the al-
gorithm is that we can easily clip a line segment against any one of the window bound-
aries. We can then perform the complete clipping by clipping against each of the four
boundaries in turn.
To clip at a boundary, we step through the drawing instructions. As we consider
each new endpoint, we decide whether it belongs to a line which crosses the boundary.
If it does, the point of intersection is determined and is passed on to the next routine.
Then each point is examined to see whether it lies within the boundary. If so, it is also
passed to the next routine. In this procedure, all line-segment endpoints lying within
the boundary and all points where lines intersect the boundary are passed on, while
points lying outside the boundary are filtered out. (See Figure 6-13.)
We can think of the process as clipping the entire figure against each window
boundary before moving on to the next boundary. (See Figure 6-14.) However, since
our clipping process steps sequentially through the figure-drawing instructions, it is
possible to begin clipping on a second boundary before the clipping of the entire figure
against the first boundary is completed. In fact, each point may be run through all four
clipping routines and entered into the display file before the next point is considered.
Algorithms for clipping a figure against each of the four window boundaries are
given below. They all follow the same outline. They first check to see if the new point
is the first point of a polygon, and if so, they save it. This is used in closing polygons
and is discussed below. They examine the new point and last point to see whether the
line segment with these endpoints crosses the boundary. The algorithms are called for
each new point. We can picture this as pen movements. We start with the pen at some
location (the last point, which is stored for each clipping boundary in the arrays XS
and YS) and ask to move it to some new position (the new point, X and Y). The clip-
ping routine examines this path to see if it encounters the clipping boundary. If it does,

arr

Filter out
Pass these these points
points

=
tee
Clipping FIGURE 6-13
boundary Clipping against an edge.
 WINDOWING AND CLIPPING 185

Clip left

Clip right

Clip bottom

Clip top

FIGURE 6-14
Clipping against all four window boundaries.

the pen is moved only to the boundary; a new command, corresponding to the clipped
point, is entered. If the side is drawn from outside the window to inside the window or
the command is for character drawing, then we introduce a MOVE command; other-
wise, the command is the same as the original. This means that if our figure passes out-
side of the window boundary, the pen will move along the boundary to the point where
the figure reenters the window region. The algorithms update the last point to be the
current point and check the current point to see whether it is inside the window. If so,
this instruction is also entered. When we say a command is ‘“entered,’’ we mean that it
is passed on to the next routine. For the first three clipping algorithms, the next routine
is the algorithm for clipping along the next boundary. The last clipping algorithm actu-
ally enters commands into the display file. (See Figure 6-15.)
186 CHAPTER SIX

Viewing transform
FIGURE 6-15
The clipping process.

6.6 Algorithm CLIP-LEFT(OP, X, Y) Routine for clipping against the left boundary
Arguments OP, X, Y a display-file instruction
Global WXL window left boundary
XS, YS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[1] THEN
BEGIN
FIRSTOP[1] < OP;
FIRSTX[1] <— X;
FIRST Yk ve
NEEDFIRST[1] < FALSE;
END
Case of drawing from outside in Ys
ELSE IF X = WXL AND XS[1] < WXL THEN
CLIP-RIGHT(1, WXL, (Y — YS[1]) * (WXL — X) / x = XS[1]) + Y)
Case of drawing from inside out
ELSE IF X = WXL AND XS[1] > WXL THEN
IF OP > 0 THEN
CLIP-RIGHT(OP, WXL, (Y — YS[1]) * (WXL — X)/
Ce eXS iE) ae 9)
ELSE
CLIP-RIGHT(1, WXL, (Y — YS[1]) « Was a os),
(X— XS[1]) + Y);
Remember point to serve as one of the endpoints of next line segment
XS[1] — x;
YSHME—Y;
 WINDOWING AND CLIPPING 187

Case of point inside

IF X = WXL AND CLOSING # 1 THEN CLIP-RIGHT(OP, X, Y);
RETURN;
END;

The calculation which occurs inside the calls to CLIP-RIGHT in the above
routine is the determination of the y coordinate of the point where the line intersects
the window boundary. The x coordinate of this point is the window boundary position.

6.7 Algorithm CLIP-RIGHT(OP, X, Y) Routine for clipping against the right bound-
ary
Arguments OP, X, Y a display-file instruction
Global WXH window right boundary
XS, YS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[2] THEN
BEGIN
FIRSTOP[2] < OP;
FIRSTX[2] <— X;
FIRST ¥[2 ¥;
e NEEDFIRST[2] <— FALSE;
END
ELSE IF X = WXH AND XS[2] > WXH THEN
CLIP-BOTTOM(1, WXH, (Y — YS[2]) *(WXH — X)/(X — XS[2]) + Y)
ELSE IF X = WXH AND XS[2] < WXH THEN
IF OP > 0 THEN
CLIP-BOTTOM(OP, WXH, (Y — YS[2]) * (WXH — X)/
Oe eXS12)) EY)
ELSE
CLIP-BOTTOM(1, WXH, (Y — YS[2]) * (WXH — X)/
(X — XS[2]) + Y);
DES
VSR Y:
IF X < WXH AND CLOSING # 2 THEN CLIP-BOTTOM(OP, X, Y);
RETURN;
END;

6.8 Algorithm CLIP-BOTTOM(OP, X, Y) Routine for clipping against the lower

boundary
Arguments OP, X, Ya display-file instruction
Global WYL window lower boundary
XS, YS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[3] THEN
188 cHAPTER SIX

BEGIN
FIRSTOP[3] < OP;
FIRSTX[3] <— X; : ‘
FIRSTY([3] <— Y;
NEEDFIRST[3] <— FALSE;
END
ELSE IF Y = WYL AND YS[3] < WYL THEN
CLIP-TOP(1, (X — XS[3]) * (WYL — Y)/(Y — YS[3]) + X, WYL)
ELSE IF Y = WYL AND YS[3] > WYL THEN
IF OP > 0 THEN
CLIP-TOP(OP, (X — XS[3]) * (WYL — Y)/
(Y = YS[3)s Xe WYP)
ELSE
CLIP-TOP(1, (X — XS[3]) * (WYL — Y)/
(Y — YS[3]) + X, WYL);
XS[3] <— X;
YS([3]) —Y;
IFY = WYL AND CLOSING # 3 THEN CLIP-TOP(OP, X, Y);
RETURN;
END;
6.9 Algorithm CLIP-TOP(OP, X, Y) Routine for clipping against the upper boundary
Arguments OP, X, Y a display-file instruction
Global WYH window upper boundary
XS, YS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[4] THEN
BEGIN
FIRSTOP[4] <— OP;
FIRSTX[4] <— X;
FIRSTY[4] — Y;
NEEDFIRST[4] <— FALSE;
END
ELSE IF Y = WYH AND YS/[4] > WYH THEN
SAVE-CLIPPED-POINT(1, (X — XS[4]) * (WYH = Y)/(Y — YS[4]) +
X, WYH) :
ELSE IF Y = WYH AND YS[4] < WYH THEN
IF OP > 0 THEN
SAVE-CLIPPED-POINT(OP, (X — XS[4]) * (WYH — Y)/
(Y — YS[4]) + X, WYH)
ELSE
SAVE-CLIPPED-POINT(1, (X — XS[4]) * (WYH — Y)/
(Y — YS[4]) + X, WYH);
XS[4] <— X;
YS[4] — Y;
IFY = WYH AND CLOSING # 4 THEN SAVE-CLIPPED-POINT(OP, X, Y);
RETURN;
END;
 WINDOWING AND CLIPPING 189

The SAVE-CLIPPED-POINT routine is used to enter the commands into the dis-
play file. It will be described below.
Let’s walk through an example to see how these routines operate. Consider the
window and sequence of line segments shown in Figure 6-16. Suppose we start with
the pen at (2, 2) and attempt to draw the lines to (4,2), (4, 4), (2, 4), and back to
(2, 2). The XS and YS array values will be initialized to the current position XS[i] = 2
and YS[i] = 2. The CLIP-LEFT routine is entered with the point X = 4, Y = 2, and
the work begins. The CLIP-LEFT routine will compare the segment from (2, 2) to
(4, 2) against the window boundary WXL = |. The segment will not require clipping
at this boundary. XS[1] is set to 4 and YS[1] is set again at 2. This point is then passed
to the CLIP-RIGHT routine. It compares the segment against the window boundary
WXH = 3. The third IF statement in this algorithm discovers that clipping is required
and passes the point X = 3, Y = 2 to the CLIP-BOTTOM routine. The XS[2] and
YS[2] values are set to 4 and 2, respectively. The CLIP-BOTTOM and CLIP-TOP
routines do not have to clip but just pass along the point and remember the (3, 2) posi-
tion in their XS, YS array elements. The command to draw the line from (2, 2) to (3, 2)
is entered into the display file by SAVE-CLIPPED-POINT. The next line segment is
seen by CLIP-LEFT as going from (4, 2) to (4, 4). Since this does not cross the left
boundary, the point (4, 4) is passed along and remembered. The CLIP-RIGHT routine
will also consider the line from (4, 2) to (4, 4). Since both points are outside the right
window boundary, this routine will not pass along the point to CLIP-BOTTOM. It will
only remember the point (4, 4) as its current pen position. The next point is (2, 4).
Again the CLIP-LEFT routine will remember the (2, 4) position in the XS[1], YS[1]
array elements and pass the point to clip right. In CLIP-RIGHT the second IF state-
ment will realize that this line crosses from outside the right boundary to inside. It will
send a command to MOVE to the point (3, 4) to the CLIP-BOTTOM routine. It re-
members the point (2, 4) and finally passes a LINE command to this point. The first of
these two calls to CLIP-BOTTOM is for the move from (3, 2) to (3, 4). This does not
cross the bottom boundary, so the routine remembers the current position (3, 4) and
passes the move command to CLIP-TOP. The CLIP-TOP routine will clip this com-
mand at (3, 3) and enter this move into the display file. It will set its current position to
(3, 4). Now back to the second call to CLIP-BOTTOM by CLIP-RIGHT. This isa line

Oa A (A)

(183) : (B73)

(4, 2)
(2, 2)

FIGURE 6-16
qd, 1) Window (3, 1) A clipping example.
190 CHAPTER SIX

command to the point (2, 4). CLIP-BOTTOM will remember this point and pass the
command to CLIP-TOP. CLIP-TOP will also remember the point, but will not pass the
command any further because the line from (3, 4) to (2, 4) is above the window. Fi-
nally, we give the point (2, 2) to the CLIP-LEFT routine. It passes the command to
CLIP-RIGHT, which passes it to CLIP-BOTTOM, which passes it in turn to CLIP-
TOP. The CLIP-TOP routine sees the segment from (2, 4) to (2, 2). The second IF
statement in the routine forwards a command to MOVE to the point (2, 3) to the dis-
play file. The final IF statement sends the command to draw a line to the point (2, 2).
The net result has been a line from (2, 2) to (3, 2), a move to (3, 3), a move to (2, 3),
and a line back to (2, 2).

THE CLIPPING OF POLYGONS

We would like our clipping routine to handle polygons as well as line segments. What
will happen if a polygon crosses our window boundary? Our clipping routine will re-
move some of the polygon’s sides, and it will insert a move command instead of a line
command along the window boundary. This change in the number of sides in the poly-
gon must be reflected in our initial polygon-drawing operation code. We will consider
the move command to be an invisible side since it occupies one instruction and moves
the pen just as a line-drawing command. Because of the change in the number of sides,
we cannot know what polygon command to enter (if any at all) until the entire polygon
has been clipped. We will therefore not enter polygon instructions into the display file
immediately. Instead, we shall store them in a temporary area. When all sides have
been clipped, we can count how many sides remain, form an appropriate polygon com-
mand, and then enter this new command (along with the instructions that were saved
for the sides) into the display file. The instructions which survive the clipping routines
are therefore treated in two different ways. Instructions which do not belong to a poly-
gon are given a viewing transformation and placed in the display file, while instruc-
tions which are part of a polygon are placed in a temporary storage buffer. This deci-
sion is made in the algorithm SAVE-CLIPPED-POINT based on a flag PFLAG which
indicates polygon processing. The algorithm also keeps track of how many polygon
sides have been saved. (See Figure 6-17.)

6.10 Algorithm SAVE-CLIPPED-POINT(OP, X, Y) Saves eee polygons in the T

buffer and sends lines and characters to the display file
Arguments OP, X, Y a display-file instruction
Global COUNT-OUT a counter of number of sides on olisbad polygon
PFLAG indicates if a polygon is being clipped
BEGIN
IF PFLAG THEN
BEGIN
COUNT-OUT <— COUNT-OUT + 1;
PUT-IN-T(OP, X, Y, COUNT-OUT);
END
ELSE VIEWING-TRANSFORM(OP, X, Y);
RETURN;
END;
 WINDOWING AND CLIPPING 191

Is it
a polygon
9

PUT-IN-T VIEWING-TRANSFORM

Saves sides in Saves sides in

T buffer display file

FIGURE 6-17
SAVE-CLIPPED-POINT.

We have called the arrays providing temporary storage for polygons IT, XT, and
YT. They must be large enough to hold the maximum number of polygon sides. The
above algorithm uses the routine PUT-IN-T to save instructions in these arrays. The al-
gorithm for PUT-IN-T is as follows:

6.11 Algorithm PUT-IN-T(OP, X, Y, INDEX) Save an instruction in the T buffer

Arguments OP, X, Y the instruction to be stored
INDEX the position at which to store it
Globai IT, XT, YT arrays for temporary storage of polygon sides
BEGIN
IT[INDEX] < OP;
XT[INDEX] < X;
YT[INDEX] < Y;
RETURN;
- END;

For polygons, we have an overall clipping routine which counts how many sides
of the original polygon have been considered. When all sides are considered, we need
to make sure that the polygon is closed. The closing problem is illustrated in Figure
6-18. Clipping against the left boundary results in a starting point above the bottom
boundary and an ending point which is below this boundary. Now clipping this se-
quence of points against the bottom boundary results in a polygon which is not closed
because there is no command to move across the boundary, and intersection points are
only calculated when the boundary is crossed.
To fix this problem we require each clipping stage to close its version of the poly-
gon. (See Figure 6-19.) To do this each clipping stage stores the first instruction it re-
ceives from the polygon. This is done by the first IF statement in each of algorithms
6.6 through 6.9. The NEEDFIRST flag is used to tell if the instruction is the first.
192 cHapTER SIX

iE eaeaerer
Le k 1
| |
|
| es ’

|
ii VS as
Fi
7

Clipping all sides...

Fi
4.
vA
Z
7
We
7
... gives unclosed polygon

“4
bp.
4
fo ben
7 FIGURE 6-18
Correct Closing the polygon.

After all commands have been sent we set the CLOSING variable and output the saved
instruction to each clipping stage. This causes each stage to check the edge between
the last point of the polygon and the first point for intersection with the clipping boun-
dary. If it does intersect, the intersection point is entered which completes the polygon.
For this final check we want to enter the intersection point, but we do not want to enter
the first point of the polygon a second time. This is what the CLOSING variable is for.
It prevents the reentry of the first point when closing the polygon. After closing the
polygon, the final number of sides of the clipped polygon is checked to see whether it
is greater than 3. If it is less than 3, then the polygon has collapsed or has been clipped
away and no entry at all should be made. If the new polygon has-an acceptable number
of sides, then we must update the polygon command to reflect this. We must also enter
the x and y coordinates of this command so that we begin drawing the polygon at the
point where drawing of the sides will terminate. All of this is done by the algorithm
CLIP-POLYGON-EDGE.

6.12 Algorithm CLIP-POLYGON-EDGE(OP, X, Y) Close and enter a clipped poly-

gon into the display file
Arguments OP, X, Y a display-file instruction
 WINDOWING AND CLIPPING 193

!
I
CLIP-LEFT receives the original polygon.
!
I ere

It is clipped and closed

to send these points to
CLIP-RIGHT,
A
|

\
Ben

A which forwards them

toe te CLIP-BOTTOM.

Clipping and closing yields A

this polygon for CLIP-TOP,
!
0 si

! i which is then saved.

lige a

FIGURE 6-19
Each clipping stage closes its polygon.

Global PFLAG indicates that a polygon is being drawn

COUNT-IN the number of sides remaining to be processed
COUNT-OUT the number of sides to be entered in the display file
IT, XT, YT temporary storage arrays for a polygon
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY arrays for saving the first command
CLOSING indicates the stage in polygon
Local I for stepping through the polygon sides

BEGIN
COUNT-IN <— COUNT-IN — 1;
CLIP-LEFT(OP, X, Y);
194 CHAPTER SIX

IF COUNT-IN # 0 THEN RETURN;

close the clipped polygon
CLOSING < 1; .
IF NOT NEEDFIRST[1] THEN CLIP-LEFT(FIRSTOP{1], FIRSTX[1], FIRSTY[1));
CLOSING < 2;
IF NOT NEEDFIRST[2] THEN CLIP-RIGHT(FIRSTOP[2], FIRSTX[2],
FIRSTY[2]);
CLOSING < 3;
IF NOT NEEDFIRST[3] THEN CLIP-BOTTOM(FIRSTOP([3], FIRSTX[3],
FIRSTY ([3]);
CLOSING < 4;
IF NOT NEEDFIRST[4] THEN CLIP-TOP(FIRSTOP[4], FIRSTX[4], FIRSTY[4]);
CLOSING < 0;

PFLAG < FALSE;

IF COUNT-OUT < 3 THEN RETURN;

enter the polygon into the display file
VIEWING-TRANSFORM(COUNT-OUT, XT[COUNT-OUT], YT[COUNT-OUT));
FOR I = 1 TO COUNT-OUT DO VIEWING-TRANSFORM(TT{I], XT[J, YT[I]):
RETURN;
END;

We must catch and handle polygon commands so that when a polygon is discov-
ered, it is entered into the temporary file. Counters are set for the number of sides to be
expected and the number of sides of the result. Last-point variables for each of the clip-
ping routines are initialized, and a flag is set so that future calls to the clipping routine
will be recognized as polygon sides. This is done by the algorithm CLIP. This is the
top-level clipping routine. Basically, it decides between handling polygons and han-
dling other graphics primitives.

6.13 Algorithm CLIP(OP, X, Y) Top-level clipping routine

Arguments OP, X, Y the instruction being clipped
Global PFLAG indicates that a polygon is being processed
COUNT-IN number of polygon sides still to be iftput
COUNT-OUT number of clipped polygon sides stored
XS, YS arrays for saving the last point drawn 3
Local I for initializing the four clipping routines
BEGIN
IF PFLAG THEN CLIP-POLYGON-EDGE(OP, X, Y)
ELSE JF OP > 2 THEN
BEGIN
PFLAG <- TRUE;
COUNT-IN < OP;
COUNT-OUT < 0;
FOR I = 1TO4DO
BEGIN
XS[{I] — X;
 WINDOWING AND CLIPPING 195

YS({ <-Y;
END;
END
ELSE CLIP-LEFT(OP, X, Y);
RETURN;
END;

ADDING CLIPPING TO THE SYSTEM

The CLIP algorithm will clip, transform, and save drawing instructions in the display
file. We have only to include it as part of the display-file instruction storage process. To
do this, we modify our DISPLAY-FILE-ENTER routine. This routine will now get the
current object-space pen position and place it on the display file through the clipping
routine, transforming it to image space dimensions in the process. (See Figure 6-20.)

6.14 Algorithm DISPLAY-FILE-ENTER(OP) (Modification of algorithm 2.23) Com-

bine operation and position to form an instruction and save it in the display file
Argument OP the operation to be entered
Global DF-PEN-X, DF-PEN-Y the current pen position
BEGIN
IF OP < 1 AND OP > —32 THEN PUT-POINT(OP, 0, 0)
ELSE CLIP(OP, DF-PEN-X, DF-PEN-Y);
RETURN;
END;

We would also like to have an initialization routine which sets the boundaries of
the viewport and window to be the same as our normalized screen coordinates, that is,
from 0 to | in both the x and y directions. This makes the window and viewport trans-
formation transparent for the user who does not wish to use it.

Windowing
Display
User program and
file
clipping

Intrepreter

FIGURE 6-20
Adding windowing to the system.
196 = CHAPTER SIX

6.15 Algorithm INITIALIZE-6

Global PFLAG polygon processing flag
XS, YS position of the pens confined by clipping boundaries
NEEDFIRST array ofindicators for saving the first command
CLOSING indicates the stage in polygon
Local I for initialization of the four clipping routines
BEGIN
INITIALIZE-5
SET-VIEWPORT(0.0, 1.0, 0.0, 1.0);
SET-WINDOW(0.0, 1.0, 0.0, 1.0);
NEW-VIEW-2;
FORI = | to4 DO
BEGIN
NEEDFIRST[I] <— FALSE;
XS[I] — 0;
YS[I] — 0;
END;
CLOSING < 0;
PFLAG < FALSE;
RETURN;
END;

GENERALIZED CLIPPING
We have used four separate clipping routines, one for each boundary. But these
routines are almost identical. They differ only in their test for determining whether a
point is inside or outside the boundary. It is possible to write these routines in a more
general form, so that they will be exactly identical and information about the boundary
is passed to the routines through their parameters. In a recursive language this would
mean that instead of having four separate routines, only one routine would be needed.
This routine would be entered four times (recursively), each time with a different
boundary specified by its parameters. Furthermore, the routine can be generalized to
clip along any line (not just horizontal and vertical boundaries). This form of the al-
gorithm is not limited to clipping along rectangular windows parallel to the axis. Clip-
ping along arbitrary lines means that the window sides may be at any angle, and by re-
cursively calling the clipping algorithm as many times as needed (not just four), the
window can have more than four sides. The generalized algorithm‘in a recursive lan-
guage can be used to clip along an arbitrary convex polygon. (See Figure 6-21.)

FIGURE 6-21
A window with six clipping boundaries.
 WINDOWING AND CLIPPING 197

POSITION RELATIVE TO
AN ARBITRARY LINE
A line divides a plane into two half planes. Let us consider briefly a half-plane test to
determine on which side ofa line a point lies. Suppose we have aline specified by the
points (x;, y,) and (xX, y2). Recall from Chapter | that if a third point (x, y) is on the
line, then

(x Xo) (yy ys) = (y= Yo) (xy = Xp) (6.3)

If the left-hand side does not equal the right-hand side, then the point is not on the line.
If the left expression is greater than the right

(Kom Ka Vay AY > V5) 4X Xd (6.4)

then the point lies on one side; if it is less

(X — X2) (Yi — Yo) < (Y — Y2) (Ki — Xa) (6.5)

then the point lies on the other side. (See Figure 6-22.) The choice of which of the two
sides corresponds to the ‘‘greater than’’ case depends upon which of the two line
points is named (x,, y)).
As an example, consider the line containing points (x,, y,;) = (1, 2) and(x>, y>) =
(4, 5). The point (x, y) = (3, 4) is on the line because

: (3 -4)(2-5) =(4-5)(l - 4) (6.6)

The point (2, 5) is in the half plane above and to the left of the line

2-4(2-5>5-5(i -4 (6.7)

For this identification of the points (x,, y,) and (X2, y2), all points (x, y) which result in
the left-hand side greater than the right-hand side are above and left of the line.

FIGURE 6-22
Boundary Deciding on which side ofa line a point lies.
198 CHAPTER SIX

The point (4, 3) is below and right. As we can easily verify, it results in the “‘less
than’ relation
4-4(2-=5)<3-5Sad-4 (6.8)
This test may be used in a clipping algorithm to determine if a point lies inside or out-
side an arbitrary boundary line.
Other forms of the test are possible. We saw that another form of the line equa-
tion is rx + sy + t = 0 (Equation 1.6). Substituting the (x, y) coordinates of a point
not on the line into the left-hand side of this expression will also give a positive
number for one side of the line and a negative number for the other side. Furthermore,
we found in Equation 1.30 that for proper normalization of r, s, and t, the magnitude of
this expression is the distance of the point from the line.

MULTIPLE WINDOWING
Some systems allow the use of multiple windowing; that is, a first image is created by
one or more window transformations on the object. Then, windows are applied to this
first image to create a second image. Further windowing transformations may be done
until the desired picture is created. Every application of a window transformation al-
lows the user to slice up a portion of the picture and reposition it on the screen. Thus
multiple windowing gives the user freedom to rearrange components of the picture.
The same effect may be achieved, however, by applying a number of single-window
transformations to the object. (See Figure 6-23.)

yet
Weal
Nea
Val
\
\
\
\
\
\
\ \ \ ‘
ee ey \
\

FIGURE 6-23
Multiple windowing.
 WINDOWING AND CLIPPING 199

AN APPLICATION
One important application of computer graphics is the design of integrated circuits.
These miniature components may be produced by photographic techniques from large
drawings of their circuitry. The drawings describe the areas of the conducting,
semiconducting, and insulating materials. Certain patterns or geometries of these ma-
terials produce the individual diodes and transistors. A single integrated circuit may
have tens of thousands of transistors. Producing a correct drawing for such a complex
structure can be quite a task, and computer graphics is an invaluable aid. There is usu-
ally a great deal of regularity and repetition within the circuit structure. Our ability to
reproduce a pattern by repeated calls upon a single image-generating subroutine is
helpful here. Furthermore, the graphics program which draws the circuit may be part of
a larger program which provides some checks of the circuit’s correctness.
The full drawing of the circuit may be | to 2 meters square. If this is reduced to
the size of the designer’s terminal, the detail will be too fine and too complex to be
useful (if it can be displayed at all). What is needed is a clipping window which dis-
plays only the portion of the circuit which the designer is currently working on. A call
on our SET-WINDOW routine will provide this. If the designer should need to look at
two separate portions of the circuit at the same time, the display surface may be sepa-
rated into two viewports and the portions of the circuit selected by two windows. The
designer could specify one window-viewport pair, open a display-file segment, draw
the circuit (clipping away all but the portion of interest), close the segment, and then
repeat the process for the second portion of the circuit to be displayed. (See Figure
6-24.)

SET-WINDOW(20.0, 30.0, 40.0, 50.0);

SET-VIEWPORT(0.2, 0.8, 0.6, 1.0);
CREATE-SEGMENT(1);
DRAW-CIRCUIT;
CLOSE-SEGMENT,
SET-WINDOW(20.0, 30.0, 10.0, 20.0);
SET-VIEWPORT(0.2, 0.8, 0.0, 0.4);

FIGURE 6-24
Use of windows and viewports to examine two portions of an
integrated circuit.
200 CHAPTER SIX

CREATE-SEGMENT(2);
DRAW-CIRCUIT;
CLOSE-SEGMENT, : .

FURTHER READING
Some other clipping methods for lines may be found in [CYR78], [LIA83], [LIA84],
and [ROG85]. The Cohen-Sutherland algorithm is also presented in [FOL82]. An im-
plementation of the Cohen-Sutherland algorithm is given in [WHI86]. The Sutherland-
Hodgman algorithm is presented in [SUT74]. The Sutherland-Hodgman algorithm
requires a convex clipping region; it also results in a single polygon, even when adivi-
sion into several independent polygons might seem more natural. An alternative clip-
ping method which overcomes these problems is given by [WEI77]. It is possible to
find the intersection between a line segment and a clipping boundary by repeatedly di-
viding the segment into halves. This is particularly useful on systems which do not
support fast division. A clipping method based on this technique is given in [SPR68].
A theoretical discussion and algorithms for the general problem of deciding if a point is
inside, outside, or on the boundary of a shape are presented in [LEE77] and [TIL80].
The problem of finding the intersection of two polygons is also considered in
[ORO82], [YAM72], and [WEI80]. A formal description of transformations and clip-
ping in a hierarchical picture structure is given in [MAL78].

[CYR78] Cyrus, M., Beck, J., ‘‘Generalized Two and Three-Dimensional Clipping,’’ Computers and
Graphics, vol. 3, no. 1, pp. 23-28 (1978).
[FOL82] Foley, J. D., Van Dam, A., Fundamentals of Interactive Computer Graphics, Addison-Wesley, Read-
ing, Mass., pp. 146-149 (1982).
[LEE77] Lee, D. T., Preparata, F P., ‘‘Location of a Point in a Planar Subdivision and Its Applications,”’
SIAM Journal on Computing, vol. 6, no. 3, pp. 594-606 (1977).
[LIA83] Liang, Y., Barsky, B. A., ‘‘An Analysis and Algorithm for Polygon Clipping,’’ Communications of
the ACM, vol. 26, no. 11, pp. 868-877 (1983).
[LIA84] Liang, Y., Barsky, B. A., “‘A New Concept and Method for Line Clipping,’’ ACM Transactions on
Graphics, vol. 3, no. 1, pp. 1-22 (1984).
[MAL78] Mallgren, W. R., Shaw, A. C., ‘‘Graphical Transformations and Hierarchic Picture Structures,”’
Computer Graphics and Image Processing, vol. 8, no. 2, pp. 237-258 (1978).
[NEW75] Newman, W. M., ‘‘Instance Rectangles and Picture Structure,’ Proceedings of the Conference on
Computer Graphics, Pattern Recognition, & Data Structures, pp. 297-301, IEEE Cat. No. 75CH0981-
Ic (1975). :
[ORO82] O’Rourke, J., Chen, C., Olson, T., Naddor, D., ‘‘A New Linear Algorithm for Intersecting Con-
vex Polygons,’’ Computer Graphics and Image Processing, vol. 19, no. 4, pp. 384-391 (1982).
[ROG85] Rogers, D. F, Rybak, L. M., ‘‘A Note on an Efficient General Line-Clipping Algorithm,’’ JEEE
Computer Graphics and Applications, vol. 5, no. 1, pp. 82-86 (1985).
[SPR68] Sproull, R. FE, Sutherland, I. E., ‘‘A Clipping Divider,’’ AFIPS FJCC, vol. 33, pp. 765-776
(1968).
[SUT74] Sutherland, I. E., Hodgman, G. W., ‘‘Reentrant Polygon Clipping,’’ Communications of the ACM,
vol. 17, no. 1, pp. 32-42 (1974).
[TIL80] Tilove, R. B., ‘“‘Set Membership Classifications: A Unified Approach to Geometric
Intersection
Problems,’’ JEEE Transactions on Computers, vol. C-29, no. 10, pp. 874-883 (1980).
[WEI77] Weiler, K., Atherton, P, ‘‘Hidden Surface Removal Using Polygon Area
Sorting,’’ Computer
Graphics, vol. 11, no. 2, pp. 214-222 (1977).
 WINDOWING AND CLIPPING 201

[WEI80] Weiler, K., “‘Polygon Comparison Using a Graph Representation,’ Computer Graphics, vol. 14,
no. 3, pp. 10-18 (1980).
[WHI86] White, M. A., Reppert, R. J., ‘‘Clipping and Filling Polygons,’’ Computer Language, vol. 3, no.
5, pp. 45-58 (1986).
[YAM72] Yamin, M., ‘‘Derivation of All Figures Formed by the Intersection of Generalized Polygons,’’ Bell
System Technical Journal, vol. 51, no. 7, pp. 1595-1610 (1972).

EXERCISES
6-1 If the viewport is the entire view screen [SET-VIEWPORT (0, 1, 0, 1)], what will be the
homogeneous-coordinate viewing transformation matrix for each of the following window set-
tings?
a) SET-WINDOW (0, 1, 0, 1)
b) SET-WINDOW (0, 1, 0, 2)
c) SET-WINDOW (0, 3, 0, 1)
d) SET-WINDOW (2, 3, 0, 1)
e) SET-WINDOW (— 10, 10, —2, 2)
6-2 If the window is left at the default setting [SET-WINDOW (0, 1, 0, 1)], what will be the
homogeneous transformation matrix for each of the following viewports?
a) SET-VIEWPORT (0, 0.5, 0, 1)
b) SET-VIEWPORT (0, 1, 0, 0.3)
c) SET-VIEWPORT (0.5, 1, 0, 1)
d) SET-VIEWPORT (0, 1, 0.25, 0.75)
- e) SET-VIEWPORT (0.25, 0.75, 0.5, 1.0)
6-3 For each window-viewport pair specified below, give the viewing transformation matrix as
homogeneous coordinates.
a) SET-VIEWPORT (0, 1, 0.25, 0.75)
SET-WINDOW (0, 1, 0, 0.5)
b) SET-VIEWPORT (0, 0.5, 0, 0.5)
SET-WINDOW (0, 2, 0, 4)
c) SET-VIEWPORT (0.5, 1, 0.5, 1)
SET-WINDOW (10, 20, 0, 10)
6-4 Consider the following figure.
r

(€5;29)

(25,07) CPS; 1)

(G25;,-2) (Gr Sete)

Sketch what will appear on the screen for each of the following window and viewport settings.
a) SET-VIEWPORT (0, 1, 0, 1)
SET-WINDOW (0, 2, 0, 2)
202 cHAPTER SIX

b)ma SET-VIEWPORT (0, 0.5, 0, 0.5)

SET-WINDOW (0, 1, 0, 1)
cwa SET-VIEWPORT (0, 1, 0, 1)
SET-WINDOW (0, 0.5, 0, 0.7)
d wa SET-VIEWPORT (0.5, 1, 0.5, 1)
SET-WINDOW (0, 1, 0, 1)
e wm SET-VIEWPORT (0, 1, 0, 0.5)
SET-WINDOW (0, 1, 0.5, 1)
6-5 Think of a set of coordinate units which are convenient for some application area. Give a
window and viewport specification which allows the use of these coordinates in creating an
image.
6-6 Suppose that a window has its lower-left corner at (—2, — 1) and its upper-right corner at
(3, 2). For each of the following line segments, state whether it will be totally visible, totally
hidden, or partially visible; and for the partially visible segments, give the coordinates at which
it is clipped.
a) eC) to Gish)
b) (1, 3) to (1.6, 1)
ONC alt, Oyitol(ae 2)
d) (3, 1) to\(4e1)
e) (—2, 3) to (1, 4)
f) (=, 3) tox4s 085)
g) (2, 3) to (4, 1.5)
h) (0, 1.5) to (2.5, 0)
6-7 Show that clipping may be done in any coordinate units (e.g., the world coordinates in
which the image is created, the normalized device coordinates, or the actual device coordinates).
Show that clipping can also be easily done in coordinates where the clipping boundaries are ro-
tated multiples of 90 degrees. Discuss clipping when the clipping boundaries are rotated by arbi-
trary angles.
6-8 What is the maximum number of additional vertices which can be created when
a) An n-sided convex polygon is clipped bya line.
b) An n-sided concave polygon is clipped byaline.
c) An n-sided convex polygon is clipped by a rectangle.
d) An n-sided concave (possibly self-overlapping) polygon is clipped by a rectangle.
6-9 Show why the Sutherland-Hodgman clipping algorithm will only work for convex clipping
regions.
6-10 Devise a clipping algorithm for clipping against a line so that it returns two polygons, one
for each side of the line. :

PROGRAMMING PROBLEMS
6-1 Implement algorithms 6.1 through 6.15 to extend your graphics system to include window-
ing and clipping.
6-2 a) Construct a circle of radius 1 centered at the origin by connecting, with line segments,
points on the circle (cos A , sin A) which are separated by an angle of 7/15 or less.
b) By using a window on the circle and a line segment, construct the symbol for an AND
gate.
 WINDOWING AND CLIPPING 203

Re
c) By using three windows on the circle construct the symbol for an OR gate.

OF
6-3 Use windows on the map of Programming Problem 3-7 to draw maps of each of the
four
counties.

6-4 Consider the following (long) shaft.

r
| i
sees 100 =
204 cnapTER SIX

a) Try drawing the entire shaft on the full screen using a square window (bigger than 104)
and a viewport.
b) Use a rectangular window which is much longer than it is,high (like a box that ‘‘fits”’
the shaft) so that the length is compressed.
c) Use two windows and two viewports to display only the ends of the shaft.
6-5 Extend a graphing routine (such as suggested in Programming Problems 2-8 and 3-11) so
that the graph is automatically scaled to be as large as possible and still fit on the display. Do this
by creating the appropriate window on the graph.
6-6 Our viewing transformation includes only scales and translations, but in general any trans-
formation could be used. In particular, a rotation can give us a viewport which can be rotated on
the display. Modify the graphics system to allow the user to specify a rotated rectangular view-
port.
*6-7 Implement the Cohen-Sutherland clipping algorithm and integrate it into the system for
use when polygons are not filled.
*6-8 Generalize the Sutherland-Hodgman algorithm for clipping against a boundary so that it
will clip against any arbitrary line. Assume that the boundary is specified by four parameters x,,
Y,, X2, Y2 which are passed to the algorithm and that a point (x, y) is visible when

(KIX) Va) (Yi a) (CX)

**6-9 If your programming language allows recursion, formulate the generalized clipping al-
gorithm of Programming Problem 6-8 in such a way that clipping against all boundaries of the
window may be carried out asa series of recursive calls of a single clipping routine. Such a
routine would permit clipping against windows which are arbitrary convex polygons.
**6-10 Devise and implement a method for clipping against an arbitrary polygon. Include user
routines for setting the clipping region as well as system routines for doing the clipping. (One
way to approach this problem is to decompose the clipping polygon into convex regions and then
apply the Sutherland-Hodgman method for each region.)
 CHAPTER

SEVEN
INTERACTION

INTRODUCTION
Computer graphics gives us added dimensions for communication between the user
and the machine. Data about size, shape, position, and orientation can be presented to
the user in a natural manner. Complex organizations and relationships can be conveyed
clearly to the user. But communication should be a two-way process. It is desirable to
allow the user to respond to this information. The most common form of computer out-
put is a string of characters printed on the page or on the surface of a CRT terminal.
The corresponding form of input is also a stream of characters coming from a
keyboard. Computer graphics extends the form of output to include two-dimensional
images, lines, polygons, arcs, colors, and intensities. The graphic display can show
the positions of objects and the relationships between objects. We might ask what is
the appropriate form of input for a user’s response to these images. First consider what
sorts of responses a user might wish to make to such a display. The user may wish to
select a particular object on the display. He may wish to specify a position on the dis-
play. He may wish to alter an object’s orientation, size, or location. He may wish to
enter a new object onto the display. We find the usual keyboard character input to be
quite unnatural for these functions. The user does not wish to determine the coordi-
nates of some point relative to some reference system and then enter them into the
machine as decimal digits; he would much rather just point to the position he is in-
terested in. It is awkward to create a drawing by typing coordinates of the endpoints of
line segments or by typing instructions for moving the imaginary pen. Most users
would prefer to be able to take a real pen and move it across the screen. Some effective
methods have been developed for the input of graphic information. A user might move
a penlike stylus and see lines appear on the display as if real pen and ink were being
used. He may ‘‘attach’’ a portion of an image to the stylus and reposition it by moving

205
206 CHAPTER SEVEt

the stylus. Sca’z and orientation might. be altered by pushing levers or twisting dials. A
portion of the display may be selected by just pointing at it. In this chapter we shall
consider the devices which allow the user to respond to graphical information in a nat-
ural manner. We shall also consider some of the techniques which may be employed to
take advantage of this interaction between man and machine.

HARDWARE
Various hardware devices have been developed to enable the user to interact in this
more natural manner. These devices can be separated into two classes, locators and
selectors. Locators are devices which give position information. The computer typi-
cally receives from a locator the coordinates for a point. Using a locator we can indi-
cate a position on the screen. Selector devices are used to select a particular graphical
object. A selector may pick a particular item but provide no information about where
that item is located on the screen.
Let us first consider some locator devices. One example of a locator is a pair of
thumbwheels such as are found on the Tektronix 4010 graphics terminal. These are two
potentiometers mounted on the keyboard, which the user can adjust. One potentiome-
ter is used for the x direction, the other for the y direction. Analog-to-digital converters
change the potentiometer setting into a digital value which the computer can read. The
potentiometer settings may be read whenever desired. The two potentiometer readings
together form the coordinates of a point. To be useful, this scheme must also present
the user with information as to which point the thumbwheels are specifying. Some
feedback mechanism is needed. This may be in the form of a special screen cursor, that
is, a special marker placed on the screen at the point which is being indicated. It might
also be done by a pair of cross hairs which cross at the indicated point. As a
thumbwheel is turned, the marker or cross hair moves across the screen to show the
user which position is being read. (See Figure 7-1.)
Another locator device is a joystick. A joystick has two potentiometers, just as a
pair of thumbwheels. However, they have been attached to a single lever. Moving the

Computer

FIGURE 7-1
Thumbwheel entry of position.
 INTERACTION 207

lever forward or back changes the setting on one potentiometer. Moving it left or right
changes the setting on the other potentiometer. Thus with a joystick both x- and y-coor-
dinate positions can be simultaneously altered by the motion of a single lever. The
potentiometer settings are processed in the same manner as they are for thumbwheels.
Some joysticks may return to their zero position when released, whereas thumbwheels
remain at their last position until changed. Joysticks are inexpensive and are quite com-
mon on displays where only rough positioning is needed. (See Figure 7-2.)
Some locator devices use switches attached to wheels instead of potentiometers.
As the wheels are turned, the switches produce pulses which may be counted. The
count indicates how much a wheel has rotated. This mechanism is often found in mice
and track balls. A mouse is a palm-sized box with a ball on the bottom connected to
such wheels for the x and y directions. As the mouse is pushed across a surface, the
wheels are turned, providing distance and direction information. This can then be used
to alter the position of a cursor on the screen. A mouse may also come with one or
more buttons which may be sensed. A track ball is essentially a mouse turned upside
down. The ball which turns the wheels is large and is moved directly by the operator.
(See Figure 7-3.) There are also mice which use photocells rather than wheels and
switches to sense position. An optical mouse is moved across a surface which contains
a special grid pattern. Photocells in the bottom of the mouse sense the movement
across the grid and produce pulses to report the motion.
If we had a paper drawing or a blueprint which we wished to enter into the
machine, we would find that the joystick was not very useful. Although the joystick
could indicate a position on the screen, it could not match the screen position to the
corresponding blueprint position. For applications such as tracing we need a device
called a digitizer, or a tablet. A tablet is composed ofa flat surface and a penlike stylus
or windowlike tablet cursor. (See Figure 7-4.) The tablet is able to sense the position of
the stylus or tablet cursor on the surface. A number of different physical principles have
been employed for the sensing of the stylus. Most do not require actual contact be-
tween the stylus and the tablet surface, so that a drawing or blueprint might be placed
upon the surface and the stylus used to trace it. A feedback mechanism on the screen is

FIGURE 7-2
Joystick.
208 CHAPTER SEVEN

FIGURE 7-3
Mouse Trackball Mouse and track ball.

not as necessary for a graphics tablet as it is for a joystick because the user can look at
the tablet to see what position he is indicating. Nevertheless, if tablet entries are to be
coordinated with items already on the screen, then some form of feedback, such as a
screen cursor, is useful.
The user may not wish to have every stylus position entered into the machine.
Some of the time he may be moving the stylus about in order to position it for the next
entry. It is therefore desirable to have some means of turning the tablet off and on, so
that the computer can only read coordinate values when the user is ready. A convenient
way of doing this is to build a switch into the tip of the stylus which turns on only
when the stylus is pressed down. The user can then lift the stylus, position it, and press
down to enter a point.
An example of a selector device is a light pen. A light pen is composed of a
photocell mounted in a penlike case. (See Figure 7-5.) This pen may be pointed at the
screen on a refresh display. The pen will send a pulse whenever the phosphor below it
is illuminated. While the image on a refresh display may appear to be stable, it is in
fact blinking on and off faster than the eye can detect. This blinking is not too fast,
however, for the light pen. The light pen can easily determine the time at which the

Tablet

FIGURE 7-4
Cursor Tablet.
 INTERACTION 209

FIGURE 7-5
Light pen.

phosphor is illuminated. Since there is only one electron beam on the refresh display,
only one line segment can be drawn at a time and no two segments are drawn simul-
taneously. When the light pen senses the phosphor beneath it being illuminated, it can
interrupt the display processor’s interpreting of the display file. The processor’s instruc-
tion register tells which display-file instruction is currently being drawn. Once this in-
formation is extracted, the processor is allowed to continue its display. Thus the light
pen tells us which display-file instruction was being executed in order to illuminate the
phosphor at which it was pointing. By determining which part of the picture contained
the instruction that triggered the light pen, the machine can discover which object the
user is indicating. It is often possible to turn the interrupt mechanism on or off during
the display process and thereby select or deselect objects on the display for sensing by
the light pen.
‘A light pen is an example of an event-driven device. Unlike a locator, which can
be sampled at any time, the processor must wait on the light pen until it encounters an
illuminated phosphor on the screen. The computer must wait for this to happen before
it can obtain the desired information. A keyboard is a more familiar example of an
event-driven device. The machine must wait for a key to be pressed before it can deter-
mine what character the user wishes to send. Buttons and switches may also be avail-
able as input devices. Again, these are event-driven devices. Handling of event-driven
devices can be conducted in two different ways. The first is to enter a polling loop, that
is, a loop which checks the status of the device until an event is indicated. When an
event occurs, the loop is exited with the results of the event. (See Figure 7-6.) Reading
of terminal input is often handled this way in high-level languages. A disadvantage of
this scheme is that the processor is essentially idle while it is waiting for an event. This
would be disastrous on a device where the processor is also needed to maintain the dis-
play.
An alternative approach is to enable an interrupt which is sent by the device when
an event occurs. An interrupt is an electrical signal which causes the processor to break
the normal execution sequence and transfer control to a special interrupt-handling
routine. By using the interrupt mechanism, the processor is free to carry out some
other operation while it is waiting for an event to take place. When the event occurs,
processing is interrupted and the device which caused it is serviced. After servicing the
device, processing can continue. (See Figure 7 -7.)
The interrupt approach to event handling allows handling of events which may
occur at times other than when the processor is expecting them. Suppose that one is
 Check for
an event

Has
an event
occurred

FIGURE 7-6
A polling loop.

Disable
interrupts
save current
state of
machine

—S Get data
Transfer of
control whenever
an interrupt
occurs

Restore state
of machine
reenable
interrupts

FIGURE 7-7
An interrupt scheme for retrieving
Main program Interrupt handler input data.

210
 INTERACTION 211]

using a program which presents a display and then requests a keyboard input from the
user. The familiar user might anticipate the correct keyboard input and enter it before
the system has finished processing and presenting the display. If polling techniques are
used, this input could be lost because the computer is not ready for it. An interrupt
scheme, on the other hand, would temporarily stop processing and store the input so
that it can be reexamined at the appropriate time. The processing of events is broken
into two parts. When events occur, the associated information is entered onto an event
queue. Several events could occur before the program is ready for them, but they
would all be stored and would therefore be accessible. When information from an
event-driven device is finally needed, the event queue is searched to see what events
have transpired. If an appropriate event has occurred, the information is removed from
the queue. If no event has occurred and the queue is empty, a polling loop can still be
employed to repeatedly check the status of the event queue, instead of checking the
status of individual devices.
We may have more than one event-driven device causing interrupts. Input data
from different devices may have different sizes or forms. We may wish, therefore, not
to save this information directly in a queue, but in some other location. We can use the
queue entry to save a ‘‘pointer’’ which tells us where to find the information for the
event. Another alternative is to have a separate specialized queue for each class of de-
vice.
Seeking the information about an event is a two-step operation. The first step is
to examine the queue to determine whether the event occurred. The second step is to
recover any information which may have been stored when the event transpired. Thus
the first step is to await the event, whereas the second is to get the results.

INPUT DEVICE-HANDLING ALGORITHMS

Once again we are dealing with an area which depends heavily on the particular
hardware devices available. The graphics standards achieve device independence by
specifying what routines should be available and what they should do (but not how
they do it); a user can then depend on these features being available. The system acts
as an interface between a device-independent user program and the particular devices
available. The ‘‘insides’’ of the interface routines will depend on the devices, but the
‘*outsides’’ which the user sees will always look the same. This is great for the user
but difficult for us, because we cannot just give an algorithm. Each particular input de-
vice will have its own version of the algorithm. This will make the routines of this
chapter a bit more nebulous than previous routines. Nevertheless, we can state in gen-
eral terms what we want our algorithms to do. All graphic input may be simulated
using only a keyboard device if necessary.
For the keyboard simulation, the internal form of the routines is substantially al-
tered from that of the general case. This is because of two factors. First, input from a
keyboard via a high-level language appears to be a sampled, rather than event-driven,
mechanism. A READ may be done at any time and will always return some value. In
effect, processing is suspended until the input is obtained. So, like a sampled device,
whenever a READ occurs, a value is returned. To act as an event-driven device, the
READ statement would have to return whether or not some new information has been
 212 = CHAPTER SEVEN

placed in an input buffer. But this is.just not the case for most high-level languages.
The second area of difference is the fact that we can assume that there is only one ac-
tual device and, therefore, need not worry about problems such as what to do when
two events occur at the same time. ~
Let us consider the general forms of routines to handle input devices. We shall
consider five classes of devices. The first is the button. The button is an event-driven
device that sends an interrupt when it is pressed. It sends no information other than the
fact that it has been pressed. The second class of device is the pick. The pick is typified
by a light pen. It is an event-driven device which selects some portion of the display.
The third class is the keyboard, and the fourth is the locator. A locator is a sampled de-
vice that returns the coordinates of some position. Finally, we shall include a valuator,
which, like a locator, is a sampled device, but which returns only one value. A valuator
might be a dial or knob which can be set by the user. There is nothing to prevent a sys-
tem from having several buttons, light pens, keyboards, locators, or valuators. There-
fore, to specify a particular piece of hardware we have to indicate not only its class but
also which member of that class the device happens to be.
We assume that some mechanism is available for turning these devices logically
on or off, so that routines should be provided which allow the user to enable or disable
each device. When a device is disabled, its inputs, if any, are ignored. The user must,
therefore, enable the device before it may be used.
Let us begin our discussion of the interactive graphics routines with the enabling
and disabling of devices. Instead of enabling and disabling individual devices, we
shall simplify this process to the enabling and disabling of device classes. We shall
create a flag for each class. The setting of the flags will determine whether or not a par-
ticular class is enabled. We can give numerical names to the various classes to aid in
their specification. (See Table 7-1.) The disable routine, then, takes as its argument a
class number, performs whatever actions are needed to turn this class of device logi-
cally off, and sets the corresponding device flag to false. The enable routine similarly
takes a class number, performs any necessary actions to turn the devices logically on,
and sets the corresponding device flag to true.

7.1 Algorithm ENABLE-GROUP(CLASS) Routine to enable an input device class

Argument CLASS the code for the class to be enabled
Global BUTTON, PICK, KEYBOARD, LOCATOR, VALUATOR device flags
BEGIN mt
IF CLASS = 1 THEN .
5

TABLE 7-1
Input class numbering
—_—_—_—_____

Input device type Class number

Button
Pick
Keyboard
Locator
Valuator MP
WN
AE See Pea eae eis Sat saree
 INTERACTION 213

BEGIN
PERFORM ALL OPERATIONS NEEDED TO PERMIT INPUT FROM THE
BUTTON DEVICES;
BUTTON < TRUE;
END;
IF CLASS = 2 THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PERMIT INPUT FROM THE
PICK DEVICES;
PICK <— TRUE;
END;
IF CLASS = 3 THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PERMIT INPUT FROM THE
KEYBOARDS;
KEYBOARD < TRUE;
END;
IF CLASS = 4THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PERMIT INPUT FROM THE
LOCATOR DEVICES;
LOCATOR < TRUE;
END;
IF CLASS = 5 THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PERMIT INPUT FROM THE
VALUATOR DEVICES;
VALUATOR < TRUE;
END;
RETURN;
END;

7.2 Algorithm DISABLE-GROUP(CLASS) Routine to disable an input device class

, Argument CLASS the code for the class to be disabled
Global BUTTON, PICK, KEYBOARD, LOCATOR, VALUATOR device flags
BEGIN
IF CLASS = 1 THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PROHIBIT INPUT FROM THE
BUTTON DEVICES;
BUTTON < FALSE;
END;
IF CLASS = 2 THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PROHIBIT INPUT FROM THE
PICK DEVICES;
PICK < FALSE;
END;
IF CLASS = 3 THEN
214 CHAPTER SEVEN

BEGIN
PERFORM ALL OPERATIONS NEEDED TO PROHIBIT INPUT FROM THE
KEYBOARDS; »

KEYBOARD < FALSE;

END;
IF CLASS = 4THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PROHIBIT INPUT FROM THE
LOCATOR DEVICES;
LOCATOR < FALSE;
END;
IF CLASS = 5 THEN
BEGIN
PERFORM ALL OPERATIONS NEEDED TO PROHIBIT INPUT FROM THE
VALUATOR DEVICES;
VALUATOR < FALSE;
END;
RETURN;
END;

We can also provide the user with a single routine for disabling all input devices.

7.3 Algorithm DISABLE-ALL Routine to disable all input devices

Local CLASS for stepping through the possible classes of devices
BEGIN
FOR CLASS = 1TO5 DO DISABLE-GROUP(CLASS);
RETURN;
END;

EVENT HANDLING
In theory we picture an event-driven device as one which generates an interrupt. When
the processor detects the interrupt, it stops its current activity and services the device.
When the device has been serviced, normal processing resumes. What is involved in
servicing an interrupt? What will an algorithm to do this servicing look like? An out-
line for such an algorithm is given below. It assumes that there is a single event queue.
In general, however, there could be many queues instead of only one. Servicing the in-
terrupt means identifying which device caused the interrupt and obtaining the input
data from that device. This information is then stored on the event queue. Strings are
handled alittle differently because they may require more storage. The string may be
stored in a special string storage area. On the event queue, instead of trying to store the
entire string in a data field, we store a pointer which tells us where the string may be
found.

7.4 Algorithm EVENT This algorithm is a model for the processing of an input-device
interrupt
BEGIN
DISABLE THE PHYSICAL INTERRUPT;
to prevent interruption of the interrupt processing
 INTERACTION 215

SAVE PROCESSOR STATUS;

DETERMINE WHICH DEVICE (CLASS AND NUMBER) CAUSED THE
INTERRUPT;
IF DEVICE IS LOGICALLY ENABLED THEN
BEGIN
IF EVENT FROM A STRING INPUT DEVICE THEN
BEGIN
GET THE INPUT STRING;
ADD-STRING-Q(STRING, DATA);
END
ELSE GET THE DATA FROM THE DEVICE;
ADD-EVENT-Q(CLASS, NUMBER, DATA);
END;
RESTORE PROCESSOR STATUS;
REENABLE PHYSICAL INTERRUPT;
RETURN;
END;

A queue is a first-in—first-out data structure like a ticket line. The first one in line
is the first to get a ticket. As new patrons arrive, they take their place at the rear of the
line. We want our algorithm to add new events to the rear of the queue. We set up ar-
rays to hold the queue data and a pointer QREAR which tells us at which point the last
entry was made. We increment this position to store the next event (if we step past the
end of the array, we wrap around to the first array position). We also have a pointer to
the next element to be removed from the queue, QFRONT. A special value of
QFRONT = 0 means that the queue is empty, so this should be checked and set to 1
when we enter the first queue element. (See Figure 7-8.)

7.5 Algorithm ADD-EVENT-Q(CLASS, NUMBER, DATA) Adds an event to the

event queue
Arguments CLASS class of the input device
NUMBER number of the input device
DATA data from the input device
“Global EVENTQC, EVENTQN, EVENTQD
arrays of size QSIZE which form the event queue
QFRONT, QREAR pointers to front and rear of event queue
Constant QSIZE the maximum size of the event queue
BEGIN
IF QREAR = QSIZE THEN QREAR < 1 ELSE QREAR < QREAR + 1;
IF QFRONT = QREAR THEN RETURN ERROR ‘EVENT Q OVERFLOW’;

FIGURE 7-8
QFRONT QREAR
An array used as a queue containing
Next to Most recently
be retrieved stored 7, 5, 8, 2, and 4.
216 CHAPTER SEVEN

EVENTQC[QREAR] < CLASS;

EVENTQN[QREAR] < NUMBER;
EVENTQD[QREAR] < DATA;
IF QFRONT = 0 THEN QFRONT < 1;
RETURN;
END;

The string queue is handled in a simpler fashion. Each new string is added to an
array of strings. When the end of the array is reached, we wrap around to the begin-
ning.

7.6 Algorithm ADD-STRING-Q(STRING, DATA) Saves STRING and returns a

pointer to it in DATA
Arguments STRING astring to be saved in the string array
DATA for return of the index of the stored string
Global STRINGQ an array of strings
SQREAR next free string storage area
Constant SQSIZE the size of the string queue array
BEGIN
IF SQREAR = SQSIZE THEN SQREAR < 1 ELSE SQREAR <— SQREAR + 1;
STRINGQ[SQREAR] < STRING;
DATA <— SQREAR;
RETURN;
END;

The user must be able to determine if an event has taken place. We must there-
fore have an ‘‘await-event’’ routine. This routine checks the event queue and returns
information about which device has caused an event. If no event has taken place and
the queue is empty, the routine may poll the queue for a specified period of time. If at
the end of this time the queue is still empty, a failure indicator may be returned.
Now let us consider the routine for checking the event queue. In the interrupt
case, this routine polls the queue until the event has occurred or time runs out. It then
obtains from the queue the class and number of the device which caused the event.
However, if the input device does not generate interrupts, then the event process
should be simulated through polling the device directly. There are two ways that the
polling may be done. It may be built into the programming language and/or operating
system we are using. Input will then act as if from a sampled device; an example is a
high-level language READ statement. Polling may also be done in a loop which we
have written ourselves. We call this a polled device. In general, our AWAIT-EVENT
routine may involve a combination of these techniques. For example, picks may be
done through true interrupts, but buttons might be detected with a ‘‘homemade’’ poll-
ing loop, and keyboard input might be obtained by a simple READ statement. In the
upcoming algorithm we give examples of all three methods. The actual implementa-
tion should include only those portions appropriate to the available hardware.

7.7 Algorithm AWAIT-EVENT(WAIT, CLASS, DEVICE) Routine to check the

event queue
Arguments WAIT the time to wait for an event to occur
 INTERACTION 217

CLASS, DEVICE to return the type of event which occurred

Global BUTTON, PICK, KEYBOARD device enabled flags
INPUT-STRING, PICKED-SEGMENT storage for keyboard and pick input
DETECTABLE segment detectability attribute array
BUTTON-POLL, PICK-POLL, KEYBOARD-POLL flag indicating
the status of polled devices
Local TIME-LIMIT the time at which to stop waiting
DATA for receiving data from the event queue
BEGIN
if buttons are simulated by a sampled device, then include the following conditional
statement
IF BUTTON THEN
BEGIN
READ DEVICE;
CLASSi<l3
RETURN;
END;
if picks are simulated by a sampled device, then include the following conditional
statement
IF PICK THEN
BEGIN
PICKED-SEGMENT < 0;
WHILE PICKED-SEGMENT < | OR PICKED-SEGMENT > NUMBER-OF-
SEGMENTS OR NOT DETECTABLE(PICKED-SEGMENT) DO
READ PICKED-SEGMENT,
CLASS*#=2;
DEVICE — 1;
RETURN;
END;
if the keyboard is simulated by a sampled device, then include the following condi-
tional statement
IF KEYBOARD THEN
BEGIN
READ INPUT-STRING;
CLASS < 3;
DEMIGE <—i1:;
RETURN;
END;
if interrupt-generating or polled devices are available, then include the following loop
TIME-LIMIT <— TIME( ) + WAIT;
WHILE TIME( ) S TIME-LIMIT DO
BEGIN
if interrupt-generating devices are adding to the event queue, then check the
queue with the following statements
BEGIN
GETQ(CLASS, DEVICE, DATA);
IF CLASS 4 0 THEN
IF CLASS =2 THEN PICKED-SEGMENT<DATA
ELSE IF CLASS = 3 THEN INPUT-STRING <— STRINGQ[DATA];
RETURN;
218 CHAPTER SEVEN

END; ;
if buttons are simulated on a polled device, then include the following condi-
tional statement
IF BUTTON AND BUTTON-POLL THEN
BEGIN
GEASS all
READ-BUTTON(DEVICE);
RETURN;
END;
if a pick is simulated on a polled device, then include the following conditional
statement
IF PICK AND PICK-POLL THEN
BEGIN
READ-PICK(DEVICE,PICKED-SEGMENT);
IF DETECTABLE[PICKED-SEGMENT] THEN
BEGIN
CLASS <2:
RETURN;
END;
END;
if the keyboard is treated as a polled device, then include the following condi-
tional statement
IF KEYBOARD AND KEYBOARD-POLL THEN
BEGIN
CLASS < 3;
READ-KEYBOARD(DEVICE, INPUT-STRING);
RETURN;
END;
END;
CLASS <0;
DEVICE < 0;
RETURN;
END;

If interrupts and a true event queue are used, then we must have a routine to get
the foremost item in queue. The algorithm uses the pointer QFRONT to indicate the
position of the leading element. If QFRONT is zero, the queue is empty; if not, the
value of the leading item is returned and the QFRONT pointer is advanced to the next
entry. ‘

7.8 Algorithm GETQ(CLASS, DEVICE, DATA) Returns the event at the front of the
event queue
If the queue is empty, zero is returned
Arguments CLASS class of the event
DEVICE device of the event
DATA input data from the event
Global EVENTQC, EVENTQN, EVENTQD the event queue arrays
QFRONT, QREAR pointers to front and rear of event queue
 INTERACTION 219

Constant QSIZE the size of the event queue

BEGIN
CLASS <0;
IF QFRONT = 0 THEN RETURN;
CLASS <— EVENTQC[QFRONT];
DEVICE <— EVENTQN[QFRONT];
DATA <— EVENTQD[QFRONT];
IF QFRONT = QREAR THEN
BEGIN
QFRONT < 0;
QREAR < 0;
END
ELSE IF QFRONT = QSIZE THEN QFRONT < 1
ELSE QFRONT <— QFRONT + 1;
RETURN;
END;

The event queue allows several events to occur before the information is used.
The queue provides storage for the event information until it is needed. But we may
sometimes want to start fresh, ignoring old events which have not been processed. We
can design a routine which will clear the event queue, discarding the unwanted events.

7.9 Algorithm FLUSH-ALL-EVENTS Removes all events from event queue

Global QFRONT, QREAR event queue front and rear pointers
SQREAR string file pointer
BEGIN
QFRONT < 0;
QREAR < 0;
SQREAR < 1;
RETURN;
END;

The AWAIT-EVENT routine simulates the occurrence of some event. For key-
board or pick events, we still must retrieve the information saved when the event oc-
curred. We must therefore provide the user with a GET-KEYBOARD-DATA routine.
This routine returns the string which was input, along with its length. A GET-PICK-
DATA routine returns the name of the segment which the user selected.

7.10 Algorithm GET-KEYBOARD-DATA(STRING, LEN) Routine to return the

stored keyboard input
Arguments STRING for the return of the string
LEN the string’s length
Global INPUT-STRING keyboard input storage
BEGIN
LEN <— LENGTH(INPUT-STRING);
STRING < INPUT-STRING;
RETURN;
END;
220 CHAPTER SEVEN

7.11 Algorithm GET-PICK-DATA(SEGMENT-NAME) Routine to return the selected

pick value
Argument SEGMENT-NAME the name of the selected segment
Global PICKED-SEGMENT pick input storage
BEGIN
SEGMENT-NAME < PICKED-SEGMENT;
RETURN;
END;

The AWAIT-EVENT routine allows several input devices to be in use at the same
time. Many applications, however, require input from only one device at a time, and
furthermore, some time-sharing systems prohibit the simultaneous use of several de-
vices. For these situations, the generality of the event queue becomes a needless over-
head. To avoid this, we can provide routines which await input from only a single class
of device, either button, pick, or keyboard. Once again, the routines will be device-
and system-dependent, but outlines for possible implementations for button and pick
routines are given below.

7.12 Algorithm AWAIT-BUTTON(WAIT, BUTTON-NUM) User routine to await the

pressing of a button
Arguments WAIT the time to wait for a button event
BUTTON-NUM for return of the number of the button device
Global BUTTON device enabled flag
BUTTON-POLL status flag if button is a polled device
Local TIME-LIMIT the time at which to stop waiting
DUMMY a dummy argument
BEGIN
IF NOT BUTTON THEN RETURN ERROR ‘BUTTON NOT ENABLED’;
if buttons are simulated by a sampled device, then include the following statement
READ BUTTON-NUM;
if interrupt-generating or polled devices are used, then include the following loop
TIME-LIMIT <—TIME( ) + WAIT;
WHILE TIME( ) = TIME-LIMIT DO
BEGIN
if interrupt-generating buttons are used, they may be found by
BEGIN
GETQ(CLASS, BUTTON-NUM, DUMMY);
IF CLASS = 1 THEN RETURN;
END;
if buttons are simulated on a polled device, then include the following condi-
tional statement
IF BUTTON-POLL THEN
BEGIN
READ-BUTTON(BUTTON-NUM);
RETURN;
END;
END;
BUTTON-NUM < 0;
 INTERACTION 22]

RETURN;
END;

7.13 Algorithm AWAIT-PICK(WAIT, PICK-NUM) User routine to await a pick

Arguments WAIT the time to wait for a pick event
PICK-NUM for return of the number of the picked segment
Global PICK device enabled flag
PICK-POLL status flag if pick is a polled device
Local TIME-LIMIT the time at which to stop waiting
BEGIN
IF NOT PICK THEN RETURN ERROR ‘PICK NOT ENABLED’:
if picks are simulated by a sampled device, then include the following statement
READ PICK-NUM;
if interrupt-generating or polled devices are used, then include the following loop
TIME-LIMIT <—TIME( ) + WAIT;
WHILE TIME( ) S TIME-LIMIT DO
BEGIN
if interrupt-generating picks are used, they may be found by
BEGIN
GETQ(CLASS, DUMMY, PICK-NUM);
IF CLASS = 2 THEN RETURN;
END;
if picks are simulated on a polled device, then include the following conditional
statement
IF PICK-POLL THEN
BEGIN
READ-PICK(PICK-NUM);
RETURN;
END;
END;
PICK-NUM < 0;
RETURN;
END;

SAMPLED DEVICES
The locator is a sampled device and, therefore, may be read at any time; there is no
need to wait for an event to be placed on the event queue. We need only a routine
which reads the locator and returns its x, y values.
Since locators are sampled devices, our simulation of a locator need not involve
the AWAIT-EVENT routine. For an actual locator device, the current coordinate values
are read from the device and returned. In a keyboard simulation, coordinate values will
be read from the keyboard and returned.

7.14 Algorithm READ-LOCATOR(X, Y)

Global LOCATOR the locator enabled flag
BEGIN
IF NOT LOCATOR THEN RETURN ERROR ‘LOCATOR NOT ENABLED’;
222 CHAPTER SEVEN

OBTAIN (X, Y) FROM LOCATOR DEVICE OR ITS SIMULATION;

CONVERT TO NORMALIZED DEVICE COORDINATES IF NECESSARY;
RETURN; } i
END;

Valuators are treated in essentially the same way as locators—one variable is re-
turned instead of a coordinate pair.

7.15 Algorithm READ-VALUATOR(V)

Global VALUATOR the valuator enabled flag
BEGIN
IF NOT VALUATOR THEN RETURN ERROR ‘VALUATOR NOT ENABLED’;
OBTAIN V FROM VALUATOR DEVICE OR ITS SIMULATION;
RETURN;
END;

THE DETECTABILITY ATTRIBUTE

A useful feature for pick devices is the ability to set the detectability of portions of the
display. We may wish to be able to pick an item from a subset of the items actually ap-
pearing on display. We want all other items of the display to be ignored by the pick de-
vice. This can be done by disabling the interrupt mechanism when portions of the dis-
play which we wish to ignore are being drawn. The interrupt mechanism is reenabled
when detectable items are being drawn. We would therefore like to present the user
with some routine for setting the detectability of portions of his display. We can give
segments a detectability attribute so that a segment may or may not be detectable by a
pick device. We do this by giving each segment a detectability flag. We provide the
user with a routine to set this flag to on or off (the default value would be off). The de-
tectability flag may be checked by the display-generating routines in a vector refresh
device to disable or enable light-pen interrupts. For other simulations the flag can still
be used to determine whether a segment is a candidate for a pick. An example is the
above AWAIT-EVENT routine for sampled pick simulations. In the case where a pick
for a particular segment is simulated by a keyboard, the detectability flag for that seg-
ment is checked by the AWAIT-EVENT routine. If the segment turns out not to be de-
tectable, then another input is requested. If the segment is detectable, then its name is
saved and AWAIT-EVENT returns, indicating that a valid pick has dccurred.
If actual interrupts are used for picks, then the MAKE-PICTURE-CURRENT or
INTERPRET routine should be extended to include enabling and disabling of inter-
rupts according to the detectability flag for the segment being interpreted.
The following algorithm allows the user to set the detectability attribute.

7.16 Algorithm SET-DETECTABILITY(SEGMENT-NAME, ON-OFF) User routine

to set the detectability attribute
Arguments SEGMENT-NAME the display-file segment being set
ON-OFF the detectability setting
Global DETECTABLE the detectability attribute array
Constant © NUMBER-OF-SEGMENTS the size of the DETECTABLE array
 INTERACTION 223

BEGIN
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT’:
DETECTABLE[SEGMENT-NAME] < ON-OFF;
RETURN;
END;

We have added a new segment attribute. This means that our routines which
manage the segment table must be extended to include this new property. There are
two routines which must be changed: the CREATE-SEGMENT routine, which ini-
tializes all attributes, and the RENAME-SEGMENT routine, which copies all attri-
butes to a new position in the table. The modified versions of these routines are given
below.

7.17 Algorithm CREATE-SEGMENT(SEGMENT-NAME) (Modification of algo-

rithm 6.4) User routine to create a named segment
Argument SEGMENT-NAME the segment name
Global NOW-OPEN the segment currently open
FREE the index of the next free display-file cell
SEGMENT-START, SEGMENT-SIZE, VISIBILITY
ANGLE, SCALE-X, SCALE-Y, TRANSLATE-X, TRANSLATE-Y
DETECTABLE the segment-table arrays
-Constant NUMBER-OF-SEGMENTS size of the segment table
BEGIN
IF NOW-OPEN > 0 THEN RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAMB’;
IF SEGMENT-SIZE[SEGMENT-NAME] > 0 THEN
RETURN ERROR ‘SEGMENT ALREADY EXISTS’;
NEW-VIEW-2
SEGMENT-START[SEGMENT-NAME] < FREE;
SEGMENT-SIZE[SEGMENT-NAME] < 0;
: VISIBILITY [SEGMENT-NAME] < VISIBILITY [0];
ANGLE[SEGMENT-NAME] <— ANGLE[0];
SCALE-X[SEGMENT-NAME] <— SCALE-X[0];
SCALE-Y[SEGMENT-NAME] <— SCALE-Y(0];
TRANSLATE-X[SEGMENT-NAME] <— TRANSLATE-X[0];
TRANSLATE-Y [SEGMENT-NAME] < TRANSLATE-Y[0];
DETECTABLE[SEGMENT-NAME] <- DETECTABLE/0];
NOW-OPEN <- SEGMENT-NAME;
RETURN;
END;

7.18 Algorithm RENAME-SEGMENT(SEGMENT-NAME-OLD, SEGMENT-

NAME-NEW) (Modification of algorithm 5.5) User routine to rename SEGMENT-
NAME-OLD to be SEGMENT-NAME-NEW
Arguments SEGMENT-NAME-OLD old name of segment
SEGMENT-NAME-NEW new name of segment
224 CHAPTER SEVEN

Global SEGMENT-START, SEGMENT-SIZE, VISIBILITY

ANGLE, SCALE-X, SCALE-Y, TRANSLATE-X, TRANSLATE-Y
DETECTABLE the segment-table arrays .
NOW-OPEN the segment currently open
Constant | NUMBER-OF-SEGMENTS the size of the segment table
BEGIN
IF SEGMENT-NAME-OLD < 1 OR SEGMENT-NAME-NEW <1
OR SEGMENT-NAME-OLD > NUMBER-OF-SEGMENTS
OR SEGMENT-NAME-NEW > NUMBER-OF-SEGMENTS THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
IF SEGMENT-NAME-OLD = NOW-OPEN OR
SEGMENT-NAME-NEW = NOW-OPEN THEN
RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-SIZE[SEGMENT-NAME-NEW] # 0 THEN
RETURN ERROR ‘SEGMENT ALREADY EXISTS’;
copy the old segment-table entry into the new position
SEGMENT-START[SEGMENT-NAME-NEW]
<— SEGMENT-START[SEGMENT-NAME-OLD];
SEGMENT-SIZE[SEGMENT-NAME-NEW]
< SEGMENT-SIZE[SEGMENT-NAME-OLD};
VISIBILITY[SEGMENT-NAME-NEW] < VISIBILITY[SEGMENT-NAME-OLD];
ANGLE[SEGMENT-NAME-NEW] <- ANGLE[SEGMENT-NAME-OLD];
SCALE-X[SEGMENT-NAME-NEW] <- SCALE-X[SEGMENT-NAME-OLD)];
SCALE-Y [SEGMENT-NAME-NEW] <- SCALE-Y[SEGMENT-NAME-OLD];
TRANSLATE-X[SEGMENT-NAME-NEW]
< TRANSLATE-X[SEGMENT-NAME-OLD];
TRANSLATE-Y[SEGMENT-NAME-NEW]
<— TRANSLATE-Y[SEGMENT-NAME-OLD)];
DETECTABLE[SEGMENT-NAME-NEW]
<— DETECTABLE[SEGMENT-NAME-OLD];
delete the old segment
SEGMENT-SIZE[SEGMENT-NAME-OLD] < 0;
RETURN;
END;

As in previous chapters, we shall provide the user with an initialization routine.

This routine sets the default values so that no device is enabled, no > events are in the
event queue, and no segments are detectable.

7.19 Algorithm INITIALIZE-7 Initialization

Global DETECTABLE the detectability attribute array
BEGIN
INITIALIZE-6;
DISABLE-ALL;
FLUSH-EVENT-Q;
DETECTABLE/[0] < FALSE;
RETURN;
END;
 INTERACTION 225

SIMULATING A LOCATOR WITH A PICK

While the light pen can be used to select an object on the screen, it does not give that
object’s position. Nor can it be used to indicate a position where there is no object, as
can be done with a joystick or tablet. In order to use a light pen for position informa-
tion, a tracking cross is employed. A tracking cross is a small cross made of four or
more separate line segments. This is placed at some known position on the screen and
selected so that it is detectable by the light pen. (See Figure 7-9.)
The pen is positioned at the center of the cross. Then as the pen is moved, it will
encounter one of the cross arms. If the pen is moved slightly to the right, it will en-
counter the right arm of the cross. This information is used to move the cross slightly
to the right. After each refresh the cross will be moved until the light pen is once again
centered. (See Figure 7-10.)
A similar action is taken for all of the other arms of the cross so that the tracking
cross will follow the movements of the light pen. Since the position of the center of the
tracking cross is known, positional information can be entered by ‘‘grabbing’’ the
cross with the light pen and moving it to the desired location.

SIMULATING A PICK WITH A LOCATOR

Suppose we have a locator, but no pick device. How can we use the locator to get the
effect of a pick? We know that the user is interested in the part of the picture located at
position (x, y), but do not know which segment is being drawn there. To find out, we
shall step through the display-file instructions segment by segment as if we were draw-
ing the image. But instead of generating line and character images, we will ask how
close the lines and characters are to the locator position. If a line or a character is suffi-
ciently close, we will assume that it is the item at which the user is pointing. We will
note what segment we are considering and return it as the value of the pick selection.
How can we determine the distance between a point (x, y) and a character? We
simplify the problem by treating the character as if it were a point (X,, y,). Then we can
use the distance formula to give

Ta D = [(x — xc)? + (y — ye)7]'”" (7.1)

If this distance is close enough
D < APERTURE Cis2)
then we have found the segment. (See Figure 7-11.)

FIGURE 7-9
Light pen and tracking cross.
226 CHAPTER SEVEN

NX FIGURE 7-10
If the light pen encounters one of the arms of the cross, the position
of the cross is shifted.

This formula will select a character if its point lies within a circle of radius
APERTURE centered on the locator point. But this formula involves multiplications
which are computationally costly. A more efficient test is to select a point lying within
a square with side 2 * APERTURE centered on the locator point. In this case, the test
condition becomes
\(x — x,)] + l(y - y,)| < APERTURE (7.3)
Absolute values are much easier to calculate than squares. (See Figure 7-12.)

APERTURE

(x, y) CREE YGg)

(x-x,)? + (y-y,)? > (APERTURE)?

Not selected

(x= B42 See

x)" + (y-y,)?< (APERTURE) 2
FIGURE 7-11
Selected Selection of a point.
 INTERACTION 227

(x-x,)? + (y-y,)? <(APERTURE)*

ke—2 APERTURE—>

® e
(x, Y) (Xe, Ye)

FIGURE 7-12
Ix-X_l + ly -y.|< APERTURE A square test area results in a more efficient test.

- We can also find the distance between a point and a line segment. We already
found an expression for this in Equation 1.30 which is

D = |rx + sy + [ (7.4)
where (x, y) is the locator point, and r, s, and t are parameters which describe the line
(see Equation 1.6) and also satisfy the normalization condition (Equation 1.7). The
values for r, s, and t satisfying these constraints for a line with endpoints (x,, y;), (X2,
y2) are
= lI (ya— y,)/d
s(x, = xiv (eS)

t = (xy; — X, y2/d
where

d= [Gs A) Os — Yo
With a little algebra, Equations 7.4 and 7.5 can be distilled into the formula

a |(x = *)G2 9) Sey ar %,)|

D 7.6
[(x2 — Xi) + (Y2 — Vi oS

This measure applies to the entire line containing the line segment. A preliminary test
should be made to determine if the point lies too far beyond the segment endpoints.
That is, we can reject the segment if
228 CHAPTER SEVEN

x < min(x,, X2) — APERTURE

or X > max(X,, X2) + APERTURE (7.7)

or y < min(y,, y3) — APERTURE

or y > max(y;, y2) + APERTURE
Here again we have a number of costly multiplications because we are accepting lines
which fall within a circle about the locator point. By considering a square about the

“| |
locator point we can use the following more efficient formula. (See Figure 7-13.)

(y2eA
SE — 1)
Se(KX- X)) 9 Wy : (X2-2—X)(y
xX,)(y - Yo
X;-—X
Xq—X Ya 4

<APERTURE (7.8)
Vertical and horizontal line segments must be treated as special cases. An al-
gorithm for simulating a pick with a light pen is given below. It steps through the seg-

(x;, ¥,)

[(&-x) (v2 -y1) - Y-ypP&% -x,)]?

eG < (APERTURE) 2

iggy nals FIGURE 7-13

min (|m(x
-x,) + y, -yl,|
a A teas selecting a line segment.
 INTERACTION 229

ment table as does our MAKE-PICTURE-CURRENT routine, but considers only visi-
ble, detectable, named segments. For each segment, it steps through the display file as
does our INTERPRET routine, but instead of performing the instruction, it applies the
above tests. The first segment which satisfies a test is returned as the result. If no such
segment is found, then zero is returned.

7.20 Algorithm PICK-SEARCH (PICK, X, Y) Routine used to simulate a pick with a

locator
It indicates which segment image is at location X, Y
Arguments PICK used to return the discovered segment
X, Y position of the simulated pick
Global APERTURE sensitivity of the pick
SEGMENT-START, SEGMENT-SIZE the segment-table arrays
DETECTABLE the segment detectability attribute array
VISIBILITY the segment visibility attribute array
IMAGE-XFORM a3 X 2 array containing the image transformation
Local SEGMENT for stepping through the possible segments
INSTRUCTIONS for stepping through the display-file segments
Constant ROUNDOFF some small number greater than any round-off error
NUMBER-OF-SEGMENTS size of the segment table
BEGIN
X1 <0;
Nie 0):
- FOR SEGMENT = 1 TO NUMBER-OF-SEGMENTS DO
IF SEGMENT-SIZE[SEGMENT] # 0
AND VISIBILITY[SEGMENT] AND DETECTABLE[SEGMENT] THEN
BEGIN
BUILD-TRANSFORMATION(SEGMENT);
DO-TRANSFORMATION(X1, Y1, IMAGE-XFORM);
PICK < SEGMENT;
FOR INSTRUCTION = SEGMENT-START[SEGMENT] TO
SEGMENT-START[SEGMENT] + SEGMENT-SIZE[SEGMENT] —
1 DO
BEGIN
GET-POINT(INSTRUCTION, OP, X2, Y2);
IF OP > 0 OR OP < —31 THEN
BEGIN
DO-TRANSFORMATION(X2, Y2, IMAGE-XFORM);
IF OP < —31
AND |X — X2| + [Y — Y2| < APERTURE THEN
RETURN;
IF OP = 2 THEN
IF X > MIN(X1, X2) — APERTURE AND
X < MAX(X1, X2) + APERTURE AND
Y > MIN(Y1, Y2) — APERTURE AND
Y < MAX(Y1, Y2) + APERTURE THEN
IF |Y2 — Y1| < ROUNDOFF OR
|X2 — X1| < ROUNDOFF THEN RETURN
ELSE IF MIN(|(Y2 — Y1)*(K — X1)/(X2 — X1)
230 CHAPTER SEVEN

+ Y1 — Y|,\(K2 — X1) *(¥ — Y1)/(¥2 — Y1)

+ X1 — X})
< APERTURE THEN RETURN;
X1 <—X2;
Y1<Y2;
END;
END;
END;
PICK < 0;
RETURN;
END;

ECHOING
An important part of an interactive system is echoing. Echoing provides the user with
information about his actions. This allows the user to compare what he has done
against what he wanted to do. For keyboard input, echoing usually takes the form of
displaying the typed characters. Locators may be echoed by a screen cursor displayed
at the current locator position. This allows the user to see the current locator setting
and to relate its position to the objects on the display. (See Figure 7-14.)
A pick may be echoed by identifying the selected objects on the display. The
selected objects may be flashed or made brighter or, perhaps, altered in color. This will
allow the user to determine whether or not he has selected the intended objects. (See
Figure 7-15.)
Buttons, when used to select menu items, can be echoed by flashing or highlight-
ing the selected items. (See Figure 7-16.)
For a valuator, display of the current setting in numerical form or as a position on
a scale is possible. (See Figure 7-17.)
It is important that some form of echoing be present, no matter what form it
takes. Without it, the user feels that he is working in the dark and may never feel com-
fortable with the program.

INTERACTIVE TECHNIQUES
We should include in our discussion of graphics input some of the techniques for in-
teractively creating and modifying pictures. Let us first consider how to add new pic-
~

FIGURE 7-14
Echoing a location.
 INTERACTION 23]

FIGURE 7-15
Echoing a pick.

ture elements to the image. Point plotting gives the user the capability of selecting a par-
ticular point on the screen. This is usually done by a combination of a locator and a
button. The locator is used to tell which point the user selects. The button indicates
when the locator is correctly positioned. The algorithm to perform point plotting must
await the button event; as soon as it occurs, the locator may be read. The selection of
points can be used in many different ways. For example, it can give endpoints of line
segments, positions for text, or translations of picture segments. It occurs so often that
a system routine for it is worthwhile.

7.21 Algorithm AWAIT-BUTTON-GET-LOCATOR(WAIT, BUTTON-NUM, X, Y)

User routine to interactively select a point
-Arguments WAIT the time to wait for a button event
BUTTON-NUM for return of the user’s button selection
X, Y the point the user selects
BEGIN
AWAIT-BUTTON(WAIT, BUTTON-NUM);
READ-LOCATOR(X, Y);
RETURN;
END;

One use for the point-plotting routine is to enter line segments. The idea here is
to connect the successive points selected by the user with line segments, like a child’s
‘“connect-the-dots’’ drawing. The user can indicate whether to continue or terminate
the process by which button is pushed. (See Figure 7-18.)
A program fragment for point plotting is as follows:

FIGURE 7-16
Echoing buttons and menu selection.
232 CHAPTER SEVEN

Value is 41.3 demdeeeh—!__1__|

FIGURE 7-17
Echoing a value.

BEGIN
BUTTON-NUM < CONTINUE;
WHILE BUTTON-NUM = CONTINUE DO
BEGIN
AWAIT-BUTTON-GET-LOCATOR(WAIT, BUTTON-NUM, X, Y);
LINE-ABS-2(X, Y);
MAKE-PICTURE-CURRENT;
END;
END;

A variation of this technique is called inking. Inking makes the locator automati-
cally leave a trail of line segments the way a pen leaves a trail of ink. It does not re-

i
Plot a point

is
i
Plot a second point

Connect the points FIGURE 7-18

with a line segment Point plotting.
 INTERACTION 233

quire the user to push a button for every line segment; instead, a new segment is drawn
whenever the locator moves a sufficient distance. (See Figure 7-19.)
The following program fragment outlines the inking techniques.

BEGIN
PEN-ON <— TRUE;
READ-LOCATOR(XOLD, YOLD);
WHILE PEN-ON DO
BEGIN
READ-LOCATOR(X, Y);
IF |((X — XOLD)| + |Y — YOLD| > NO-MOVEMENT THEN
BEGIN
LINE-ABS-2(X, Y);
XOLD < X;
YOLD <-Y;
MAKE-PICTURE-CURRENT;
END;
UPDATE(PEN-ON);
END;
END;

The UPDATE(PEN-ON) determines whether inking should cease. This may be

governed by an event, such as lifting the stylus, or by the passing of a time limit or a
limit on the allowed number of line segments. Note that we could collect the sequence
of locator positions without connecting them with line segments. Such a sequence of
points is called a stroke. Inking displays strokes.
Raster displays can extend inking to painting with a brush. A brush is a pattern of
pixel values. When painting, we repeatedly sample the locator position. Each time the
position changes we copy the brush pattern into the frame buffer. The pattern is cen-
tered on the current locator position. The effect is to draw this pattern repeatedly on the
display along the path of the locator. The instances of the pattern usually overlap, re-
sulting in a shape that looks as though the brush has moved across the display leaving a
trail of paint. The shape of the trail depends on both the path of the locator and the
shape of the brush. (See Figure 7-20.)
It is often the case that the user wants to connect endpoints of several lines, and
frequently horizontal and vertical lines predominate. But it can be difficult to position
to a single point, and it may be hard to get a line perfectly horizontal or vertical. A
technique to aid in such constructions is a grid. A grid is a pattern of special points.
(See Figure 7-21.) Often the user can make these points either visible or invisible.

a FIGURE 7-19
Inking.
234 CHAPTER SEVEN

FIGURE 7-20
Brush Painting.

What is special is that values obtained from the locator are rounded to the nearest grid
point. All lines begin and end exactly on a grid point. This makes it easy to ensure that
endpoints meet and that lines are vertical or horizontal. The following program frag-
ment draws a grid; it should be’placed in its own display-file segment so that it can be
made visible or invisible.

BEGIN
FOR X = 0 TO NUMBER-OF-GRIDPOINTS DO
BEGIN
FOR Y = 0 TO NUMBER-OF-GRIDPOINTS DO
BEGIN
MOVE-ABS-2(X/NUMBER-OF-GRIDPOINTS — DX/2,
Y/NUMBER-OF-GRIDPOINTS);
LINE-REL-2(DX, 0);
MOVE-REL-2(— DX / 2, —DX/ 2);

++tett+etet+tt
ttt
t+e¢ti¢¢++e
tttttet++4+4+
t++e+tteet+t++et
t++t+t+tGe+¢et+4¢+
t+Htt+t+tt++et+4¢+
eee
eee
eet tere
ae
oe
ga
ee
Rt
t++ettttr+t+t++ttt+
eel FIGURE 7-21
ee++4t4¢4+44+4+4+4+4++4+
t+++et
ttt
t++t+ttet+t+tttt+
A grid.
 INTERACTION 235

LINE-REL-2(0, DX);
END;
END;
END;

In the above fragment, NUMBER-OF-GRIDPOINTS is a count of how many grid

points to display across the screen and DX is the width of a small cross which will be
imaged at each grid point.
The rounding to the nearest grid point would look like the following:

READ-LOCATOR(X, Y);
X <— INT(X « NUMBER-OF-GRIDPOINTS + 0.5) / NUMBER-OF-GRIDPOINTS;
Y <— INT(Y * NUMBER-OF-GRIDPOINTS + 0.5) / NUMBER-OF-GRIDPOINTS;

Vector refresh and some raster displays allow a technique called rubber band
lines. This technique allows the user to see what a line will look like before fixing it in
place. The basic process is the same as line plotting; the user adjusts the locator to
place the endpoint of the next line segment. The user presses a button when the locator
is correctly positioned. The difference is that a line is constantly drawn between the
last endpoint and the locator. As the locator moves, the line is redrawn in a new posi-
tion, giving the appearance of a rubber band stretched between the fixed endpoint and
the locator point. (See Figure 7-22.)

FIGURE 7-22
Rubber band lines.
236 CHAPTER SEVEN

This technique is difficult to implement in a straightforward manner in our sys-

tem without adding either the ability to alter portions of the display file without open-
ing and closing segments, or the ability to append new instructions onto a previously
closed segment. It can, however, be done by using the image transformation on a
single fixed line in a closed segment to create the rubber band line, and another open
segment to hold the user’s final drawing. The details of this are left as an exercise.
The user might wish not only to create new images but to alter and adjust exist-
ing images as well. The first thing necessary is often identifying which part of the
image is to be changed. This is naturally a pick operation. Suppose the user wants to
remove portions of the image. An AWAIT-PICK operation will determine which seg-
ment should be removed and the SET-VISIBILITY routine can make it invisible.
Notice that this technique cannot be used to bring the item back, since one cannot pick
what one cannot see. (See Figure 7-23.)
Another desirable modification might be a change in the image transformation.
For example, the user may wish to change the position of a segment’s image. It is usu-
ally most convenient for the user to indicate the new position by pointing at it with a
locator. The user can use a pick to select the segment to be moved and then select some
point on the image with the locator. Finally, the locator is moved to the place to where
the point on the image should be moved. (See Figure 7-24.)
A program fragment to do this is given below. INQUIRE-IMAGE-TRANSLA-
TION(TX, TY) is a routine which returns the current image translation parameters.

Select a segment

FIGURE 7-23
Make it invisible Selecting a segment with a pick.
 INTERACTION 237

Select object

Locate position

Ci
FIGURE 7-24
Move object Positioning.

BEGIN
AWAIT-PICK(WAIT, SEGMENT);
AWAIT-BUTTON-GET-LOCATOR(WAIT, BUTTON-NUM, XOLD, YOLD);
AWAIT-BUTTON-GET-LOCATOR(WAIT, BUTTON-NUM, XNEW, YNEW);
INQUIRE-IMAGE-TRANSLATION(SEGMENT, TX, TY);
SET-IMAGE-TRANSLATION(SEGMENT, TX + XNEW — XOLD,
r Ie GNE We eV OD):
MAKE-PICTURE-CURRENT;
END;

On a vector refresh and some raster displays, the image can be continuously
shifted and redrawn to follow the locator movements. The image appears to be at-
tached to the locator. The user can see how the picture will look before fixing the new
image translation values. This technique is called dragging. (See Figure 7-25.)
An outline for a possible dragging procedure is given below.

BEGIN
AWAIT-PICK(WAIT, SEGMENT);
AWAIT-BUTTON-GET-LOCATOR(WAIT, BUTTON-NUM, XOLD, YOLD);
DRAGGING < TRUE;
WHILE DRAGGING DO
238 CHAPTER SEVEN

BEGIN
INQUIRE-IMAGE-TRANSLATION(SEGMENT, TX, TY);
READ-LOCATOR(XNEW, YNEW); %
IF XNEW # XOLD OR YNEW # YOLD THEN
BEGIN |
SET-IMAGE-TRANSLATION(SEGMENT, TX + XNEW — XOLD,
TY + YNEW — YOLD);
XOLD <— XNEW;
YOLD <— YNEW;
MAKE-PICTURE-CURRENT,
END;
UPDATE(DRAGGING);

The UPDATE(DRAGGING) routine is used to stop the dragging process. This is

usually done by detecting the occurrence of some event, but might also be triggered by
the passing of a time limit.
Changes in orientation and scale may also be done interactively. The techniques
are basically the same as for changes of the translation parameters, except that the nat-
ural input device might be a valuator instead of a locator. A valuator may be simulated
using a locator and a scale drawn on the screen (usually parallel to the x or y axis). The

FIGURE 7-25
Dragging.
 INTERACTION 239

locator cursor can be positioned on the scale by the user and this position can then be
converted into a valuator value. (See Figure 7-26 and Plate 3.)
An interactive program may provide the user with several options. For example,
the user might be able to enter a single new line, ink, make a segment visible, move a
segment, or quit. The user must select which option to follow. Such a selection can be
made with the buttons by assigning a button for each option. The user makes a selec-
tion by pressing the appropriate button. We may wish to inform the user just what his
options and what the corresponding buttons are. Such a list of options is called a
menu. (See Figure 5-7 and Plate 1.) When there are a large number of options, a menu
can become excessively long and hard to read. To avoid this the menu can be broken
up into several smaller submenus. A top-level menu can be used to select the submenu
containing the desired item. Thus menus can be structured into a hierarchy. In a com-
plex selection process one might have to step through several levels of menus to reach
the final selection, but at each level the menu is small and the selection easy.
Menus should be placed in a segment or segments different from the main pic-
ture so that they can be made invisible when no longer needed. A user may find it in-
convenient to transfer his attention back and forth between buttons (or keyboard) and
locator (or pick). It’s for this reason that a light pen or tablet stylus will often have a
switch built into its tip or handle. To avoid switching back and forth between different
devices, we may wish to use the light pen or stylus to indicate the desired menu item.
Menu selections can be made by a pick instead of a button if we place each selection in
a separate segment so that it can be identified. When a light pen is the pick device,
such selection items are called light-buttons. They behave as buttons, but do not require
the user to shift his attention from the screen.

FURTHER READING
Description and classification of graphics input devices are found in [OHL78]. Early
descriptions of tablets are given in [DAV64] and [TEI68}. One input device we did not

ee ee FIGURE 7-26
Using scales.
240 cHAPTER SEVEN

e ITEM 1
e@ ITEM 2
e ITEM 3 : *
e ITEM 4

FIGURE 7-27
A menu.

mention is a touch panel, which acts as a locator attached to the display so that posi-
tions can be indicated by touching the screen with a finger or with a special pen. An
example of this class of device is described in [NAG85]. A version of a mouse which
detects changes in position optically is described in [LYO82]. An event queue for
graphical input is suggested in [SPR75]. Graphic interaction techniques are described
in [BER78], [FOL74], and [NEW68]. A detailed model of device-independent virtual
input devices is given in [ROS82]. A description of virtual input is given in [WAL76].
In some cases, the virtual devices we have described may not be a good match to the
actual devices; an alternative approach is to describe devices by the state they manage,
the events they notice, and the actions they cause [ANS79]. A discussion of some in-
teractive techniques for raster displays and encodings of raster data to support them
may be found in [BAU80]. The distinction between the handling of an interaction and
the applications program which uses the information is made clear in [KAM83]. The
use of picks in interactive graphics is described in [WEL80]. An algorithm for simulat-
ing a pick with a locator is given in [GAR80]. Simulating valuators with a locator is
described in [EVA81] and [THO79]. In [TUR84] a distinction is made between pro-
gram variables (set by assignment statements) and graphics variables (set by
hardware). An extension of graphics systems to allow attributes and coordinates to be
graphics variables allows the display to be dynamically altered for interaction. Issues
in interactive graphics and a bibliography are given in [THO83].

[ANS79] Anson, E., “‘The Semantics of Graphical Input,’’ Computer Graphics, vol. 13, no. 2, pp. 113-120
(1979).
[BAU80] Baudelaire, P., Stone, M., “‘Techniques for Interactive Raster Graphics,’’ Computer Graphics,
vol. 14, no. 3, pp. 314-320 (1980).
[BER78] Bergeron, R. D., Bono, P. R., Foley, J. D., ‘‘Graphics Programming Using the CORE System,”’
ACM Computing Surveys, vol. 10, no. 4, pp. 389-394 (1978). .
[DAV64] Davis, M. R., Ellis, T. O., “‘The Rand Tablet: A Man-Machine Graphical Communication De-
vice,’’ AFIPS FJCC, vol. 26, pp. 325-331 (1964).
[EVA81] Evans, K. B., Tanner, P. P., Wein, M., ‘‘Tablet-Based Valuators That Provide One, Two, or Three
Degrees of Freedom,’” Computer Graphics, vol. 15, no. 3, pp. 91-97 (1981).
[FOL74] Foley, J. D., Wallace, V. L., ‘‘The Art of Natural Graphic Man-Machine Conversation,’’ Proceed-
ings of the IEEE, vol. 62, no. 4, pp. 462-470 (1974).
[FOL84] Foley, J. D., Wallace, V. L., Chan, P,, ‘‘The Human Factors of Computer Graphics Interaction
Techniques,’’IEEE Computer Graphics and Applications, vol. 4, no. 11, pp. 13-48 (1984).
[GAR80] Garret, M. T., ‘‘Logical Pick Device Simulation Algorithms for the CORE System,’
Computer
Graphics, vol. 13, no. 4, pp. 301-313 (1980).
 INTERACTION 241

[KAM83] Kamran, A., Feldman, M. B., “Graphics Programming Independent of Interaction Techniques
and Styles,’’ Computer Graphics, vol. 17, no. 1, pp. 58-66 (1983).
[LYO82] Lyon, R. F, Haeberli, M. P., ‘‘Designing and Testing the Optical Mouse,’ VLS/ Design, vol. 3,
no. 1, pp. 20-30 (1982).
[NAG85] Nagayama, T., Shibuya, J., Kawakita, T., ‘‘Pen-Touch-Type Electro-Magnetic Transparent Touch
Panel,’’ SID Digest, vol. 16, pp. 32-35 (1985).
[NEW68] Newman, W. M., ‘‘A Graphical Technique for Computer Input,’’ Computer Journal, vol. 11, no.
1, pp. 63-64 (1968).
[OHL78] Ohison, M., ‘‘System Design Considerations for Graphics Input Devices,”’ Computer, vol. 11, no.
11, pp. 9-19 (1978).
[ROS82] Rosenthal, D. S. H., Michener, J. C., Pfaff, G., Kessener, R., Sabin, M., ‘‘The Detailed Seman-
tics of Graphics Input Devices,’’ Computer Graphics, vol. 16, no. 3, pp. 33-38 (1982).
[SPR75] Sproull, R. FE, Newman, W. M., ‘‘The Design of Gray-Scale Graphics Software,’’ Proceedings of
the Conference on Computer Graphics, Pattern Recognition, & Data Structure, pp. 18-20, IEEE Cat.
No. 75CH0981-1c (1975).
[TEI68] Teixeira, J. F., Sallen, R. P., ‘The Sylvania Data Tablet: A New Approach to Graphic Data Input,”’
AFIPS SJCC, vol. 32, pp. 315-321 (1968).
[THO79] Thornton, R. W., ‘‘The Number Wheel: A Tablet Based Valuator for Interactive Three-Dimensional
Positioning,’’ Computer Graphics, vol. 13, no. 2, pp. 102-107 (1979).
[THO83] Thomas, J. J., Hamlin, G., ‘Graphical Input Interaction Technique (GIIT) Workshop Summary,”’
Computer Graphics, vol. 17, no. 1, pp. 5-30 (1983).
[TUR84] Turner, J. U., ‘‘A Programmer’s Interface to Graphics Dynamics,’’ Computer Graphics, vol. 18,
no. 3, pp. 263-270 (1984).
[WAL76] Wallace, V. L., ‘“The Semantics of Graphic Input Devices,’’ Computer Graphics, vol. 10, no. 1,
pp. 61-65 (1976).
[WEL80] Weller, D. L., Carlson, E. D., Giddings, G. M., Palermo. F. P., Williams, R., Zilles, S. N.,
“Software Architecture for Graphical Interaction,’’ JBM System Journal, vol. 19, no. 3, pp. 314-330
(1980).

EXERCISES
7-1 Design a routine RUBBERBAND which repeatedly shows a line connecting the point
where the locator was positioned when the routine was entered and the current locator position.
When an event occurs, the segment which was open when the routine was entered will receive a
command to draw aline to the current locator position and the routine will be exited. Write al-
gorithms for RUBBERBAND under each of the following conditions.
a) Our system is extended by a routine REPLACE-INSTRUCTION(OP, X, Y) which re-
placés the last entered display-file instruction by its arguments.
b) Our system is extended by a routine EXTEND-SEGMENT(SEGMENT-NAME)
which reopens a segment so that additional instructions may be added.
c) Our system is not extended at all, so that the image transformation routines must be
used.
7-2 Outline an algorithm for a light-pen tracking cross.
7-3 a) Compare the time required on your machine to calculate the nearness of two points
using the measure in Equation 7.1 and that of Equation 7.3.
b) Compare the time required on your machine to calculate the nearness of a point to a
line using the measure in Equation 7.6 and that in Equation 7.8. Also compare the time required
if lines were stored in the general form and Equation 7.4 could be used.
c) Describe under what conditions the two measures used for determining nearness will
yield different decisions.
242 CHAPTER SEVEN

7-4 It can be very frustrating to make mistakes in an interactive graphics session, particularly if
one accidentally deletes a complex object. List a set of graphics editing commands you would
like to provide a user. For each command, discuss how easy it would be for the user to undo the
command. For those commands which would be difficult to undo*(e.g., delete), discuss how a
special ‘‘undo’’ operation might be included in the system.

PROGRAMMING PROBLEMS
7-1 Using algorithms 7.1 through 7.19 as a guide, extend your graphics system to include
interactive capabilities.
7-2 Write a program which will allow a user to interactively create a picture by using the locator
and buttons. Use a button value of 1 to move the pen to the locator position without drawing a
line (MOVE-ABS-2). Use a button value of 2 to mean drawaline to the locator position (LINE-
ABS-2). Use a button value of 3 to stop the program (TERMINATE).
7-3 a) Revise the line-drawing program of Programming Problem 7-2 so that locator points
are rounded to grid points for some reasonably spaced grid. Allow the user to either display or
not display the grid.
b) Allow the user to interactively change the grid spacing.
7-4 Write a program for inking. Use the pressing of buttons to start and stop the inking process.
Your program should repeatedly sample the locator position. If the locator has moved suffi-
ciently from the end of the inked line, extend the line to the current locator position by the addi-
tion of an instruction to the display file.
7-5 Revise Programming Problem 7-4 so that the angle between any two line segments is al-
ways less than a maximum angle which may be set as a parameter.
7-6 Write a program for moving an image. Select an image to be moved by a pick. When the
pick occurs, read the locator position. Then wait for a button to be pressed. When this second
event occurs, read the new locator position. Determine the change in locator position and add
this change to the current image translation for the selected segment.
7-7 a) Write a program which creates three shapes (a triangle, a square, and a cross) in three
different display-file segments. Set the visibility for these shapes to be false. Display a menu
which allows the user to make the shapes visible by the pressing of buttons.
b) Place each menu selection in a different display-file segment, so that it may be
selected by a pick instead of a button.
c) Let a pick of one of the three images itself cause that image to become invisible again.
7-8 Write a simple *‘pong’’ game where a ‘‘ball’’ bounces off four enclosing walls and a pad-
dle, which is controlled by a valuator. (There are several ways to make an object appear to
move. One is to use the image transformation, erasing and redrawing the entire object. Another
is to add pixels to the forward side of the object and to remove them from the back side. Use the
method which is best for your hardware.)
7-9 One of the problems with using a tracking cross and light pen to act as a locator is that a
quick motion of the pen can lose the tracking cross. Write a procedure which will search for the
lost light pen whenever it is lost. Your procedure should take a Starting position as an argument
(this might be the last known position of the pen). It should draw a search pattern of line seg-
ments. It stops when a line segment encounters the light pen and returns that location.
*7-10 Using image transformations on a closed segment containing a single horizontal line of
length 1, implement rubber band lines.
 INTERACTION 243

*7-11 Implement rubber band rectangles where one corner is anchored and the opposite corner
is attached to the locator. As the locator is moved, the sides of the rectangle stretch or shrink.
*7-12 Systems with frame buffers usually allow the programmer to inquire as to the value of a
pixel as well as to set it. We can use this to extract patterns from images on the display. Write a
program which allows the user to specify a rectangular boundary on the display. When the user
signals that the rectangle is correctly sized and positioned, your program should extract the pat-
tern of pixel values which the rectangle surrounds. The user should then be allowed to use this
pattern to fill polygons (possibly with transformations applied as described in Chapter 4).
*7-13 We have described the locator as returning the coordinates of a point in the normalized de-
vice coordinates. However, the user would often prefer to receive the point in the same coordi-
nates as he specified the drawing (before the image transformation). We can apply the inverse of
the image transformation to the locator position to obtain this value, but if there are several dis-
play-file segments on the screen, there can be several image transformations to choose from.
One approach to resolving this question is to first try to identify which segments have been
drawn in the area near the locator position. An algorithm such as 7.20 for identifying a segment
from a position may be used. But it is still possible to have more than one segment drawn in the
region. To further resolve the choice, the system can require the user to assign priorities to the
segments. When we must choose between segments, we select the one with the highest priority.
Design and implement such a mechanism.
*7-14 A computer graphics layout program can find uses in many applications (for example,
positioning stories and advertisements in a newspaper, or walls in a building, or equipment in a
room). Write an interactive layout program. Allow the user to draw the boundary area (e.g., the
walls of a room or the boundaries of a page). Provide grids and rubber band lines to make this
easy. Allow the user to select standard items for positioning from a menu (e.g., the pieces of
equipment). Selecting an item from the menu should cause an instance of it to be drawn on the
display. The user should then be able to drag this item around the display for positioning. The
user should also be able to drag any item. You may also wish to allow the user to scale or rotate
any item.
*7-15 Extend the multiple workstation idea of Programming Problem 2-14 to include multiple
input sources. Devise a means of receiving input from whichever workstations are currently ac-
tive. You should also be able to identify which workstation provided events or to select which
workstation will be sampled.
**7.16 Often when constructing pictures, the user will want to connect a line segments
endpoint to another endpoint. The careful positioning of the locator needed to connect the new
segment exactly on the old segment may be difficult and tedious. We can make life easier for the
user by writing a routine which returns a point right on a line segment when the locator is near to
that line segment, and right on an endpoint when the locator is near that endpoint. Using
algorithm 7.16 as a guide, write such a routine.
 CHAPTER

EIGHT )
THREE
DIMENSIONS

INTRODUCTION
Some graphics applications are two-dimensional; charts and graphs, certain maps, and
some artist’s creations may be strictly two-dimensional entities. But we live in a three-
dimensional world, and in many design applications we must deal with describing
three-dimensional objects. If the architect would like to see how the structure will actu-
ally look, then a three-dimensional model can allow him to view the structure from dif-
ferent viewpoints. The aircraft designer may wish to model the behavior of the craft
under the three-dimensional forces and stresses. Here again, a three-dimensional de-
scription is required. Some simulation applications, such as docking a spaceship or
landing an airplane, also require a three-dimensional view of the world.
In this chapter we shall generalize our system to handle models of three-dimen-
sional objects. We shall extend our transformations to allow translation and rotation in
three-dimensional space. Since the viewing surface is only two-dimensional, we must
consider ways of projecting our object onto this flat surface to form the image. Both
parallel and perspective projections will be discussed.

3D GEOMETRY
Let us begin our discussion of three dimensions by reviewing the methods of analytic
geometry for specifying objects. The simplest object is, of course, the point. As in the
case of two dimensions, we can specify a point by establishing a coordinate system
and listing the coordinates of the point. We will need an additional coordinate axis for
the third dimension (three axes in all, one for height, one for width, and athird for
depth). We can arrange the three axes to be at right angles to each other and label the

244
 THREE DIMENSIONS 245

width and height axes x and y, respectively, to correspond to the two-dimensional case.
The third axis, for depth, is named the z axis. We have a choice as to which end of the
z axis will be given positive values. (See Figure 8-1.)
The orientation of the coordinate system is determined as follows. If the thumb
of the right hand points in the positive z direction as one curls the fingers of the right
hand from x into y, then the coordinates are called a right-handed system. If, however,
the thumb points in the negative z direction then it is left-handed. Whereas most of
geometry uses a right-handed system, computer graphics often prefers the left-handed
coordinates. We shall adopt the right-handed system for our use. (The right-handed
system is normally used in mathematics, although in computer graphics we also find
the left-handed system in use so that objects behind the display screen will have posi-
tive z values.) Any point in space may now be specified by telling how far over, up,
and out it is from the origin, where the three axes intersect. These distances are written
as the ordered triple (x, y, z). (See Figure 8-2.)

y axis

x axis

a Right-handed

y axis

Z axis

X axis

FIGURE 8-1
Right-handed and left-handed coordinate

Left-handed systems.
246 CHAPTER EIGHT

y axis

x axis

FIGURE 8-2
Z axis Position of the point (x, y, Z).

We can use the coordinates to specify a line. In two dimensions a line was given
by an equation involving x and y.

Jigme yi zeae (8.1)

ae >,S) Kame a

This equation states that as x changes, the y coordinate changes proportionately. The
ratio of the change in y to the change in x is always the same. In three dimensions, as
we move along the line, both y and z coordinates will change proportionately to x. A
line in three dimensions is given by a pair of equations

Saag yo as
EK Toes
(8.2)
La Z) Zo a Z\

Ke a Xj X2 Te Xy

It requires the coordinates of two points to form these equations: (X,, y,, Z,) and (x9,
Y2, Zz). Thus it still takes two points to specify a line, as we.might expect. A more
symmetrical expression for a line is the parametric form, where each of the coordinates
is expressed in terms of a parameter u. ;
X = (Xp — X))u+ xX,

Vas eee Vie te Va (8.3)

Z (Zo = Zy)U: +24
We shall also want to work with planes. A plane is specified by a single equation of the
form.

Ax + By + Cz+D=0 (8.4)
 THREE DIMENSIONS 247

Notice that one of the constants (for example A, if it is not zero) may be divided out
of
the equation so that

x + By +Ciz+D,=0 (8.5)
is an equation for the same plane when
B, = B/A, C, =C/A, and D,=D/A (8.6)
It therefore requires only three constants B,, C,, and D, to specify the plane. The equa-
tion for a particular plane may be determined if we know the coordinates of three
points (not all in a line) which lie within it: (Xi, Yi, Z1){X2, Yo, Zp), and (x3, y3, Zs).
We can determine the equation in the following manner. Since each point is in the
plane, it must satisfy the plane’s equation

Xy a Biy, ar Ciz, ate D, =()

Xo aia Biy2 ae C,Z> ae D, =0 (8.7)

X3 i Biy3 Ate C)z3 SP D, = 0

We have three equations to solve for the three unknowns By, C,, and D,, and standard
algebra techniques for solving simultaneous equations will yield their values.
Another normalization is
‘Ap = A/d, Bs = Bid, Cie sd, ands", D> =.D/d.. (8.8)
where

d = (A? + B? +'C’)!? (8.9)

The value of this selection is that the distance between apoint (x, y, z) and the plane is
given by
iL =|A,X a Boy Te C,z Sr D,| (8.10)

The sign of the quantity in the absolute value bars indicates on which side of the plane
the point lies. This is an extension of Equation 1.30 for lines.
“ Another way of specifying a plane is by a single point in the plane and the direc-
tion perpendicular to the plane. A vector perpendicular to a plane is called a normal
vector. We can call [N, N, N,] the displacements for the normal vector, and (x,,
Yp» Zp) are the coordinates of a point in the plane. (See Figures 8-3 and 8-4.)
If (x, y, Z) is to be an arbitrary point in the plane, then [(x — x,) (y— yp) (z
— Z,)] is a vector parallel to the plane. We can derive an equation for the plane by
using the vector dot product. The dot product of two vectors is the sum of the products
of their corresponding components. For example, if A = [A, A, A,] and B =
[B, B, B,], then their dot product is
A:B =A,B, + A,B, + A,B, (8.11)

The result of the dot product is equal to the product of the lengths of the two vectors
times the cosine of the angle between them. (See Figure 8-5.)
248 CHAPTER EIGHT

FIGURE 8-3
ke——d, >| A vector in three dimensions.

A:B = |A||B|cosé (8.12)

Now the angle between any vector parallel to a plane and the normal vector to the
plane is 1/2 radians (because that is what we mean by a normal vector). Since the
cosine of 1/2 is 0, we know that the dot product of the normal vector with a vector
parallel to the plane is 0. We can find a vector parallel to the plane by taking the differ-
ence of two points within the plane. Therefore, if
N, (x — x,) + Ny (y — y,) + N,(@ — z,) = 0 (8.13)
is a true equation then the vector formed by the difference between (x, y, z) and (x,,
Yp> Zp) must be parallel to the plane. And since (x,, yp, Zp) is in the plane and we can
get to (x, y, z) by moving along a vector parallel to the plane, we know that (x, y, z) is
also a point in the plane. (See Figure 8-6.) So Equation 8.13 is then an equation for the
plane.

[N, Ny N,]

FIGURE 8-4
A vector normal to a plane.
 THREE DIMENSIONS 249

FIGURE 8-5
A The angle between two vectors.

3D PRIMITIVES
Now that we know how to express points, lines, and planes in three dimensions, let us
consider algorithms which allow the user to model three-dimensional objects. We have
provided the user with commands for moving the ‘‘pen’’ and drawing lines and poly-
gons in two dimensions. Extending these operations to three dimensions will mean re-
questing the user to provide three coordinate specifications instead of two. At some
time in the display process we must project these three coordinates onto the two screen
coordinates, but we shall place this burden on the DISPLAY-FILE-ENTER routine and
postpone its discussion for the moment. As in the case of two dimensions, we picture
an imaginary pen or pointer which the user commands to move from point to point. We
save this pen’s current position in some global variables, only now three variables are
required: DF-PEN-X, DF-PEN-Y,and DF-PEN-Z, one for each of the three coordi-
nates. The three-dimensional LINE and MOVE algorithms are as follows:

(x, y, Z)

(Xp5 Vp» Zp)

FIGURE 8-6 é
The angle between a vector normal to a plane and a vector parallel to the plane is 90 degrees.
250 CHAPTER EIGHT

move
8.1 Algorithm MOVE-ABS-3(X, Y, Z) The 3D absolute
pen to
Arguments X, Y, Z world coordinates of the point to move the
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z current pen position
BEGIN : ,
DF-PEN-X < X;
DF-PEN-Y < Y;
DF-PEN-Z < Z;
DISPLAY-FILE-ENTER(1);
RETURN;
END;

8.2 Algorithm MOVE-REL-3(DX, DY, DZ) The 3D relative move

Arguments DX, DY, DZ changes to be made to the pen position
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
BEGIN
DF-PEN-X < DF-PEN-X + DX;
DF-PEN-Y < DF-PEN-Y + DX;
DF-PEN-Z <— DF-PEN-Z + DZ;
DISPLAY-FILE-ENTER(1);
RETURN;
END;

8.3 Algorithm LINE-ABS-3(X, Y, Z) The 3D absolute line-drawing routine

Arguments X, Y, Z coordinates of the point to draw the line to
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
BEGIN
DF-PEN-X < X;
DF-PEN-Y < Y;
DF-PEN-Z < Z;
DISPLAY-FILE-ENTER(2);
RETURN;
END;

8.4 Algorithm LINE-REL-3(DX, DY, DZ) The 3D relative line-drawing routine

Arguments DX, DY, DZ displacements over which aline is to be drawn
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
BEGIN :
DF-PEN-X <— DF-PEN-X + DX;
DF-PEN-Y <— DF-PEN-Y + DY;
DF-PEN-Z < DF-PEN-Z + DZ;
DISPLAY-FILE-ENTER(2);
RETURN;
END;

Polygon commands can also be extended by simply processing an additional

array of coordinates. We assume that the user will provide coordinates which all lie in a
plane, but a check could be placed in the algorithms to guarantee it. The following are
the three-dimensional versions of our routines for entering polygons.
 THREE DIMENSIONS 251

8.5 Algorithm POLYGON-ABS-3(AX, AY, AZ, N) The 3D absolute polygon-drawing

routine
Arguments N the number of polygon sides
AX, AY, AZ arrays of the coordinates of the vertices
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
Local I for stepping through the polygon sides
BEGIN
IF N < 3 THEN RETURN ERROR ‘SIZE ERROR’;
DF-PEN-X <— AX[N];
DF-PEN-Y < AY[N];
DF-PEN-Z <— AZ[N];
DISPLAY-FILE-ENTER(N);
FORI = 1TON DO LINE-ABS-3(AX[I], AY[I], AZ[I]);
RETURN;
END;

8.6 Algorithm POLYGON-REL-3(AX, AY, AZ, N) The 3D relative polygon-drawing

routine
Arguments N the number of polygon sides
AX, AY, AZ arrays of displacements for the polygon sides
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
Local I for stepping through the polygon sides
TMPX, TMPY, TMPZ storage of point at which the polygon is closed
BEGIN
IF N < 3 THEN RETURN ERROR ‘SIZE ERROR’;
move to starting vertex
DF-PEN-X < DF-PEN-X + AX[1];
DF-PEN-Y <— DF-PEN-Y + AY{[1];
DF-PEN-Z <— DF-PEN-Z + AZ[1];
save vertex for closing the polygon
TMPX < DF-PEN-X;
TMPY < DF-PEN-Y;
TMPZ < DF-PEN-Z;
DISPLAY-FILE-ENTER(N);
enter the polygon sides
FOR I = 2TON DO LINE-REL-3(AX[I]J, AY[I], AZ[I]);
close the polygon
LINE-ABS-3(TMPX, TMPY, TMPZ);
RETURN;
END;

3D TRANSFORMATIONS
sing the above routines the user can construct a three-dimensional “‘drawing’’ of his
object. But the actual display surface is two-dimensional. The two-dimensional image
corresponds to a particular view of the three-dimensional object. The process of find-
ing which points on the flat screen correspond to the lines and surfaces of the object in-
252 CHAPTER EIGHT

volves a viewing transformation. We have seen how it is useful to translate, scale, and
rotate images. We shall begin by considering the generalization of these transforma-
tions to three dimensions. We shall then extend our transformations to include parallel
and perspective projections.
We saw that a two-dimensional scaling transformation matrix was of the form

Sx 0
0 Sy

or in homogeneous coordinates,

n” x So 0
0 n
se 0
0 So 1

Three dimensions gives us a third coordinate which can have its own scaling factor, so
we shall require a 3 X 3 matrix rather than a2 X 2.

Sx 0 0O
0 sy 0
0 O sz

If using homogeneous coordinates, a 4 X 4 matrix is required.

50:27 OO
0 SyrT OG
6-0 en (8.14)
OC Ors Ore

Transformation of a point is done by multiplication by the matrix, just as in Chapter 4.

sx "0 6" 0
my) eye es w;| = x y ZW
0 Sy 0 0
0 5. Sz 0 = Sex Sy y $z7Z WwW

0 0 0 1 (8.15)

We used the bottom row of the homogeneous coordinate matrix for translation values.
This will still be the case for three dimensions, only the matrix is now 4 x4instead
of
3 X 3. To translate by t, in the x direction, t, in the y direction, and t, in the z direc-
tion, we multiply by the matrix

ey I (8.16)
oor Oo
or=O ©
oal > oN CO
OO
-
<<
 THREE DIMENSIONS 253

When we considered rotation of an object in two dimensions, we developed a matrix

for rotation about the origin.

cos@ sing
-sin@ cos@

We can generalize this to a three-dimensional rotation about the z axis. If we rotate

about the z axis, all z coordinates remain unchanged, while the x and y coordinates be-
have the same as in the two-dimensional case. (See Figure 8-7.)
A matrix which does this in homogeneous coordinates is

cos@ sin6 0 0
-sing cosg 0 0
Rz = 0 Car (8.17)
0 ODO)

In the above rotation, we are thinking of the axes as fixed while some object in the
space is moved. We could also think in terms of the object being fixed while the axes
are moved. The difference between these two interpretations is the direction of rota-
tion. Fixing the axes and rotating the object counterclockwise is the same as fixing the
object and moving the axes clockwise.
Rotation about the x or y axis is done by a matrix of the same general form as in
Equation 8.17, as we might expect from the symmetry of the axes. To rotate about the
X axis So that y is turned into z, we use the homogeneous coordinate matrix

1 0 Grey
0 cos@ sing 0
Rx =| -sing cose 0 SdH)
0 0 0 1

y axis

X axis

FIGURE 8-7
Z axis Rotation about the z axis.
254 CHAPTER EIGHT

y axis

X axis

FIGURE 8-8
Z axis Rotation about the x axis.

(See Figure 8-8.) To rotate about the y axis so that z is turned into x, we use the
homogeneous coordinate matrix

cos@ QO -sin@ 0
0 1 O20 (8.19)
Ry =| gin 6 0 cose® 0
0 0 Oza rl
(See Figure 8-9.)
Three-dimensional transformations are useful in presenting different views of an
object. An example of a program which lets the user interactively specify the transfor-
mation parameters for viewing a molecule is shown in Plate 3. Since we will find
translation and rotation about an axis useful in creating a viewing transformation, we
will develop algorithms for creating transformation matrices involving these opera-
tions. We shall do this by means of a routine which creates an identity transformation
matrix (NEW-TRANSFORMATION-3) and routines which have the effect of multiply-
ing the current transformation matrix by a matrix for translation or rotation about one
of the principal axes. We will use the homogeneous coordinate techniques. For the
viewing transformation, we can arrange matters so that the fourth column is always
three Os and a 1. Knowing this, we do not actually need to store these numbers, so a
4 x 3 array of storage is all that is needed. We will call this transformation matrix
TMATRIX. The following algorithm sets it to the identity transformation.

8.7 Algorithm NEW-TRANSFORM-3 Initializes the viewing transformation matrix to

the identity
Global TMATRIX a 4 X 3 transformation matrix array
Local I, J for stepping through the array elements
BEGIN
FOR I = 1TO4DO
BEGIN
FOR J = 1TO 3 DOTMATRIX{[I, J] <0;
 THREE DIMENSIONS 255

y axis

X axis

FIGURE 8-9
Z axis Rotation about the y axis.

IFI ~ 4THEN TMATRIX{I, I] <1;

END;
RETURN;
END;

The next algorithm has the effect of multiplying the TMATRIX array byatrans-
lation matrix.

8.8 Algorithm TRANSLATE-3(TX, TY, TZ) Post-multiplies the viewing transforma-

tion matrix bya translation
Arguments TX, TY, TZ the amount of the translation
Global TMATRIX a 4 X 3 transformation matrix array
BEGIN
TMATRIX[4, 1] <—TMATRIX[(4, 1] + TX;
TMATRIX[4, 2] — TMATRIX[4, 2] + TY;
TMATRIX[(4, 3] <—TMATRIX(4, 3] + TZ;
RETURN;
-END;

Algorithms for rotating about each of the major axes are given below. The argu-
ments are the sine and cosine of the rotation angle rather than the angle itself.

8.9 Algorithm ROTATE-X-3(S, C) Post-multiplication for a rotation about the x axis

(y into z)
Arguments _ S, C the sine and cosine of the rotation angle
Global TMATRIX a 4 X 3 transformation matrix array
Local I for stepping through the matrix elements
TMP temporary storage
BEGIN
FOR I = 1TO4DO
BEGIN
TMP < TMATRIX{I, 2] * C — TMATRIX[I, 3] « S;
TMATRIX[I, 3] <TMATRIX[I, 2] « S + TMATRIX[I, 3] * C;
256 CHAPTER EIGHT

TMATRIX[(I, 2] < TMP;

END;
RETURN;
END;

8.10 Algorithm ROTATE-Y-3(S, C) Post-multiplication for a rotation about the y axis

(z into x)
Arguments S, C the sine and cosine of the angle of rotation
Global TMATRIX a 4 X 3 transformation matrix array
Local I for stepping through the matrix elements
TMP temporary storage
BEGIN
FOR I = 1TO4DO
BEGIN
TMP <—TMATRIX[I, 1] * C + TMATRIX[I, 3] * S;
TMATRIX(I, 3] <— —TMATRIX[I, 1] * S + TMATRIX[I, 3] « C;
TMATRIX[I, 1] <—TMP;
END;
RETURN;
END;

8.11 Algorithm ROTATE-Z-3(S, C) Post-multiplication for a rotation about the z axis

(x into y)
Arguments S, C the sine and cosine of the angle of rotation
Global TMATRIX a4 3transformation matrix array
Local I for stepping through the matrix elements
TMP temporary storage
BEGIN
FOR I = 1TO4DO
BEGIN
TMP <—TMATRIX[I, 1] * C — TMATRIX[I, 2] « S;
TMATRIX[I, 2] <TMATRIX[I, 1] * S + TMATRIX[I, 2] « C;
TMATRIX[I, 1] <— TMP;
END;
RETURN;
END;

ROTATION ABOUT AN ARBITRARY AXIS

Let us take a moment to become more familiar with three-dimensional transformations
by solving a problem. While we have developed transformations for rotation about a
coordinate axis, in general any line in space can serve as an axis for rotation. The prob-
lem is to derive a transformation matrix for a rotation of angle @ about an arbitrary
line. We will build this transformation out of those which we have already encoun-
tered. We shall perform a translation to move the origin onto the line. We shall then
make rotations about the x and y axes to align the z axis with the line. The rotation
about the line then becomes arotation about the z axis. Finally, we apply inverse trans-
formations for the rotations about the y and x axes and for the translation to restore the
line and coordinates to their original orientation. (See Figure 8-10.)
 THREE DIMENSIONS 257
¥ y

Zz
xX x x

zi z

(a) (b) (c)

Z Zz

Zz

x x

(@) (e)

ate
r (g) (A)
FIGURE 8-10
(a) To rotate about an arbitrary axis, (b) first translate the axis to the origin, (c) rotate about x until the axis of
rotation is in the xz plane, (d) rotate about y until the z axis corresponds to the axis of rotation, (e) rotate
about z (the axis of rotation) instead of fixing the object and rotating the axes (we have in this figure fixed
the axes and rotated the object), (f) reverse the rotation about y, (g) reverse the rotation about x, and (h) re-
verse the translation.

We should decide upon a convenient representation for the line which is to be the
axis of rotation. A point on the line, together with the line’s direction, is sufficient to
specify the line. This will prove to be a good form for our purposes. The point provides
information for the translation, and the direction will tell us the correct angles of rota-
tion for aligning the z axis. We can find this form from the parametric equations for the
line in the following manner. Given the line
 258 CHAPTER EIGHT

x = Au
+ X;

y=Bur+y, (8.20)
Zo ACUetayZi

a point on the line is (x;, y;,Z,) and the direction is specified by the vector of [A B C].
We can now determine the transformation matrix needed for our general rotation.
The initial translation to move the origin to the rotation axis should be

0
0 (8.21)
1
“x”
oo ae>) = 2. _ oO
OO
-

After this translation, the point on the line (x,, y;, Z;) will be moved to the origin. We
will also require the inverse of this translation to place the origin back at its original po-
sition once our rotations are completed. This is just a translation by the same amount in
the opposite direction

On 107-50
Tan ; ; ; (8.22)
iD A
x~*OOrF Z; 1

The next step in the process is a rotation about the x axis. We wish to rotate until our
’ general rotation axis lies in the xz plane. To determine the necessary angle of rotation,
let us place our direction vector at the new origin and consider its projection in the yz
plane. We can picture this in the following way: The line segment between (0, 0, 0)
and (A, B, C) will be in the direction of the arbitrary rotation axis, and since we have
translated so that the point (0, 0, 0) is on this axis, the entire segment must lie along
the axis. Now imagine shining a beam of light parallel with the x axis and looking at
the shadow of the line segment formed on the yz plane. The shadow will extend from
(0, 0, 0) to (0, B, C). (See Figure 8-11.)

y axis

(O, B,C) (A, B, C)

ieee
Whe

X axis FIGURE 8-11

Projection of a line segment onto the yz
Z axis plane.
 THREE DIMENSIONS 259

If we now rotate about the x axis until the arbitrary axis is in the xz plane, the
line segment’s shadow will lie along the z axis. How large an angle I is needed to place
this shadow on the z axis? (See Figure 8-12.)
We know that the length of the shadow V will be
V = (B? +°C%12 (8.23)
and from the definition of the sine and cosine, we see that
sinI = B/V
(8.24)
cosI = C/V
The rotation about the x axis should be

1 0 0 0
0 cosI sind Q
Rx = 0 -sinI cosI QO (8.25)
0 0 0 1

So we have

1 0 0 0
Ree Oo CIN BEY: 0 (8.26)
Zo ehO SBI YC hVS0
0 0 0 1

The inverse transformation is a rotation of equal magnitude in the opposite direction.

Reversing the direction of the angle changes the sign of the sine elements, but leaves
the cosine elements unchanged.

1 0 0 O
Rote DCL Biv ~O (8.27)
OTBIN. Ci V0
oO 0 0 1

y axis

FIGURE 8-12
Z axis Parameters of the line segment projection.
260 CHAPTER EIGHT

Now we can picture our rotation axis as lying in the xz plane. (See Figure 8-13.)
The rotation about the x axis has left the x coordinate unchanged. We also know
that the overall length of the segment
l= & B+ Cc)” (8.28)
is unchanged. The z coordinate will be
(L? — A)? = (B? + C7 = V (8.29)
We wish to rotate by an angle J about the y axis so that the line segment aligns with the
Z axis.
sinJ = A/L
(8.30)
cos J = V/L

The rotation matrix will be

cosJ 0 sinJ 0O
Ry = 0 1 0 0
-sinjJ OQ cosJ QO
0 0 0 1
(8.31)
V/L 0O A/L O
= 0 1 0 O
-A/L 0 V/L 0
0 0 0 1

The signs are different in Equation 8.31 from those in Equation 8.19 because we are
rotating from x into z instead of from z into x. The inverse for this transformation is

V/L 0O -A/L 0O
Ry-! — 0 1 Oe. 26 8.32
y AL 0 Vib 0 Oa
0 0 0 1

y axis

A :
x axis

Vv

Z axis Rotation axis

FIGURE 8-13
The rotation axis lying within the xz plane.
 THREE DIMENSIONS 261

Finally we are in a position to rotate by an angle 0 about the arbitrary axis. Since
we have aligned the arbitrary axis with the z axis, a rotation by @ about the z axis is
needed.

cos@ sino 0 QO
Fagot -sing@ cos@ 0 QO
Zz 0 0 1 0 (8.33)

0 0 On

The actual transformation for a rotation 6 about an arbitrary axis is given by the prod-
uct of the above transformations.
Ro = TR,RyR,Rj'R;'T" (8.34)

PARALLEL PROJECTION
We have talked about creating and transforming three-dimensional objects, but since
our viewing surface is only two-dimensional, we must have some way of projecting our
three-dimensional object onto the two-dimensional screen. (See Figure 8-14.)
Perhaps the simplest way of doing this is just to discard the z coordinate. This is
a special case of a method known as parallel projection. A parallel projection is formed
by extending parallel lines from each vertex on the object until they intersect the plane

y axis

X axis

3 D object

Zz axis

y axis

FIGURE 8-14
X axis A three-dimensional object and its
Projection of object projection.
262 CHAPTER EIGHT

of the screen. The point of intersection is the projection of the vertex. We connect the
projected vertices by line segments which correspond to connections on the original
object. (See Figure 8-15.)
Our special case of discarding the z coordinate is the case where the screen, or
viewing surface, is parallel to the xy plane, and the lines of projection are parallel to
the z axis. As we move along these lines of projection, only the z coordinate changes;
x and y values remain constant. So the point of intersection with the viewing surface
has the same x and y coordinates as does the vertex on the object. The projected image
is formed from the x and y coordinates, and the z value is discarded.
In a general parallel projection, we may select any direction for the lines of pro-
jection (so long as they do not run parallel to the plane of the viewing surface). Sup-
pose that the direction of projection is given by the vector [x, y, Zp] and that the
image is to be projected onto the xy plane. If we have a point on the object at (x), yj,
Z,), we wish to determine where the projected point (x2, y>) will lie. Let us begin by
writing the equations for a line passing through the point (x, y, z) and in the direction
of projection. This is easy to do using the parametric form
Ao Xe Ape

y=yY¥1 + ypu (8.35)

DSL oe

y axis,

Z axis

X axis

y axis

H
xaxis FIGURE 8-15
Projection A parallel projection.
 THREE DIMENSIONS 263

Now we ask, where does this line intersect the xy plane? That is, what are the x and y
values when z is 0? If z is 0, the third equation tells us that the parameter u is

Z}
u= —-— (8.36)

Substituting this into the first two equations gives

Xo = X; — Z (x,/Z,)
8.37
Ve Vi — 21 (nt 2) wall
This projection formula is in fact a transformation which may be written in matrix
form

1 0
[X2 yo] =(0€x1 yi: 21] 0 1 (8.38)
-Xp/Zp ~Yp/zp

or in full homogeneous coordinates

1 0 OF,
- [Xe y2 22 1) re
=([x1 y: Z 1) 0
ees 1
ype 06 06 (8.39)
0 0 Oz

This transformation always gives 0 for z,, the z position of the view plane, but it is
often useful to maintain the z information. We will find it useful in Chapter 9 when we
order projected objects according to their z position as part of the hidden-surface re-
moval process. A transformation that includes determining a z-coordinate value z,
(which turns out to be the same as z,) is as follows:

. 1 0 0 0
~ 0 1 0.20
[X2 y2 22 1]
= (x; yi % 1) -xp/Zp -Yp/Zp 1 0 (8.40)
0 0 1

We will just ignore the z value when drawing the projected image.
Let us write an algorithm for performing parallel projections. We shall assume
that the ratios

SXP = x,/Z,
(8.41)
SYP = y,/Z,

have been calculated and stored, so as to avoid recomputing them for every point pro-
jected.
264 CHAPTER EIGHT

8.12 Algorithm PARALLEL-TRANSFORM(X, Y, Z) Parallel projection of a point

Arguments X, Y, Z the point to be projected, also for return of result
Global SXP, SYP the parallel projection vector ratios|
BEGIN
X <—X — Z x SXP;
Y<Y —Z* SYP;
RETURN;
END;

PERSPECTIVE PROJECTION
An alternative projection procedure is a perspective projection. In a perspective projec-
tion, the further away an object is from the viewer, the smaller it appears. This pro-
vides the viewer with a depth cue, an indication of which portions of the image corre-
spond to parts of the object which are close or far away. In a perspective projection, the
lines of projection are not parallel. Instead, they all converge at a single point called
the center of projection. It is the intersections of these converging lines with the plane of
the screen that determine the projected image. The projection gives the image which
would be seen if the viewer’s eye were located at the center of projection. The lines of
projection would correspond to the paths of the light rays coming from the object to the
eye. (See Figure 8-16.)

y axis

Z axis

X axis

y axis

x FIGURE 8-16
Projection A perspective projection.
 THREE DIMENSIONS 265

If the center of projection is at (x,, y,, Z.) and the point on the object is (x;, y;,
Z,), then the projection ray will be the line containing these points and will be given by

Rye Kote (ky ku

ve aa You (8.42)
Ree Pat (2) > Zoi
The projected point (x2, y) will be the point where this line intersects the xy plane.
The third equation tells us that u, for this intersection point (z = 0), is

u=- (8.43)
Z) Zo

Substituting into the first two equations gives

X Xo
kp ee
Z) Zo
(8.44)
y1 Yc
SPE ae
Z;— Z
With alittle algebra, we can rewrite this as
XZ} — XZ
2 XS
Z}— Z
(8.45)
Yc4%1 — YiZe
2 ae teat Aaa
Z;— Z

This projection can be put into the form of a transformation matrix if we take full ad-
vantage of the properties of homogeneous coordinates. The form of the matrix is

“Zc O O 0
OZ CoO
r Pitts aVe Our 1 (3:46)
0 0> 0 >=Zc

To show that this transformation works, consider the point (x,, y;, Z,). In homoge-
neous coordinates we would have

[Xiw, YW, ZW, Wi]

Multiplying by the transformation matrix gives

-Ze¢ 0 0 0
0c. "Zc. 10-0
[ X2W2 Yo2W2 Z2W2 W2)] =[xXiw, yiwi ziW; Wi] Re. YeoO. 1
ORG ri S26

= [-x1wizc + ZiwixXc -yiwiZc + Ziwiyc O Ziwi-zcwi] (8.47)

266 CHAPTER EIGHT

sO

W2 = Z,W; — ZW,

and

LN
(8.49)

which gives
Zo = 0 (8.50)

and
XyW2 = —X,WyZ, + ZW Xe (8.51)

which gives

X2 =
XZ} — X1Z
(8.52)
Z; — Z

and

Y2W2 = —YiWiZ, + ZW (8.53)

which gives

YZ) — YiZe
= (8.54)
Z; — Z,

The resulting point (x, y2) is then indeed the correctly projected point.
An equivalent form of the projection transformation is

1 0 0 0
0 1 0 0 (8.55)
Py ™xe/ze, -Yyo/zc. 0 -1/zZe
0 0 0 1

The change is the factor of — 1/z,. Because we divide the first three coordinates (x,,,
Yw> Zw) by w to obtain the actual position, changing all four coordinates by some com-
mon factor has no net effect. In the literature, the perspective transformation is often
defined such that the center of projection is located at the origin and the view plane is
positioned at z = d. For this situation, the transformation is given by

(8.56)
—oOo
~
© Qa

O71©
oO
So oO
of SSCs
 THREE DIMENSIONS 267

We can show that the descriptions are equivalent as follows. To get the perspective
transformation with the center of projection at (x,, y,, Z.), we can first translate this
point to the origin, and then use the perspective transformation matrix P, given in
Equation 8.55, where d is —z,. We finally translate back to the original coordinates.
The details of the proof are left as an exercise.
The above transformation maps points on an object to points on the view plane.
We shall, however, find it useful to use a modified version of this transformation. We
want a form which yields the same x and y values so that we can still display the
image, but we would like to compute a z value different from zero. We would like to
find z values such that we can preserve the depth relationship between objects, even
after they are transformed. If object A lies in front of object B, we want the perspective
transformation of object A to lie in front of the perspective transformation of object B.
Furthermore, we want to compute z values such that after transforming the points on a
straight line, we still have a straight line. We want the transformation to preserve plan-
arity so that polygons are not warped into nonpolygons. The reason we want to do this
is to allow us to establish depth order for hidden-surface removal (in Chapter 9) after
having performed the perspective transformation. A form of the perspective transfor-
mation which meets these requirements is as follows:

eZ.Cate eg AO ae
Peace (8.57)
; P=! xX Yo -1 1
0 0 0 -Z¢

This yields a transformed z value given by

Zy
13 (8.58)
Zj he Zo

We shall now present an algorithm for performing a perspective transformation

of a point.
“8.13 Algorithm PERSPECTIVE-TRANSFORM(X, Y, Z) For perspective projection
of a point
Arguments X, Y, Z the view plane coordinates of the point
Global XC, YC, ZC the center of projection
Local D the denominator in the calculations
Constant | ROUNDOFF some small number greater than any round-off error
VERY-LARGE a very large number approximating infinity
BEGIN
Dic Ze AZ,
IF |D] < ROUNDOFF THEN
BEGIN
X <—(X — XC) * VERY-LARGE
Y <—(Y — YC) * VERY-LARGE
Z <— VERY-LARGE
END
268 CHAPTER EIGHT

ELSE
BEGIN
X — (KX * ZC — XC
* Z)/D;
Y<(Y¥* ZC — YC *Z)/D;
Z<—Z/D;
END;
RETURN;
END;

VIEWING PARAMETERS
We have seen how parallel and perspective projections can be used to form a two-di-
mensional image from a three-dimensional object as seen from the front. But suppose
we wish to view the object from the side, or the top, or even from behind. How can
this be done? All that we need to do is to apply some rotation transformations before
projecting. There are two equivalent ways of thinking about this. We can think of the
view plane (that is, the plane of our display surface) as fixed and the object as rotated,
or we can picture the object as fixed and the view plane as repositioned. (See Figure
8-17.)
In our system we shall use this second description. It is as if the view plane were
the film in a camera. Every display-file segment represents a photograph taken by this
camera. (See Figure 8-18.)
We can move the camera anywhere, so we can view the object from any angle.
The picture taken by this synthetic camera is what is shown on the display surface, as
if the film is developed and stuck upon the screen. The user is given routines by which
he may change a number of viewing parameters. By setting the parameters, he can po-
sition the synthetic camera.(See Figure 8-19.)
The first parameters we shall consider are the coordinates of the view reference
point (XR, YR, ZR). The view reference point is the center of attention. All other view-
ing parameters are expressed relative to this point. If we rotate the view, it will be a ro-
tation about the view reference point (not about the origin). We can think of the view
reference point as an anchor to which we have tied a string. The synthetic camera is at-
tached to the other end of the string. By changing other viewing parameters, we can
swing the camera through an arc or change the length of the ee One end of the
string is always attached to the view reference point.
The direction of this imaginary string is given by the view ‘plane normal vector
[DXN DYN DZN]. This normal vector is the direction perpendicular to the view
plane (the view plane is the film in the camera). This means that the camera always
looks along the string toward the view reference point. The camera is pointed in the di-
rection of the view plane normal.
The length of the string is given by the VIEW-DISTANCE parameter. This tells
how far the camera is positioned from the view reference point. The view plane is
positioned VIEW-DISTANCE away from the view reference point in the direction of
the view plane normal.
We now have two coordinate systems. We have the object coordinates which we
used to model our object, and we have view plane coordinates, which are attached to the
 THREE DIMENSIONS 269

y axis

X axis

y axis

Axes fixed
Object moved

Z axis

SO

X axis ee,

y axis

Object fixed
ae Axes moved

z axis- Ss FIGURE 8-17

Rotating an object is equivalent to
rotating the axes in the opposite direc-
Xx axis tion.

view plane. Think of the object coordinates as being painted on the ‘‘floor’’ which sup-
ports the object, while the view plane coordinates have been printed on the film in the
synthetic camera. We can place the origin of the view plane coordinates at the point
where the string attaches to the film, that is, where a line parallel to the view plane nor-
mal, passing through the view reference point, intersects the view plane.
There is one more viewing parameter we need to discuss. Imagine holding the
string fixed and spinning the camera on it so that the string serves as the axis of rota-
tion. At each angle, the photograph will show the same scene, but rotated so that a dif-
ferent part of the object is up. (See Figure 8-20.)
270 CHAPTER EIGHT

FIGURE 8-18
The synthetic camera analogy.

The view-up direction [XUP YUP ZUP] fixes the camera angle. Imagine an
arrow extending from the view reference point in the view-up direction. We look
through the camera’s viewfinder and spin the camera until the arrow appears to be in
the camera’s “‘up’’ direction.
Changing the view reference point will change the part of the object that is
shown at the origin. (See Figure 8-21.)

View reference
point

View plane
normal

View plane

FIGURE 8-19
Three-dimensional viewing parameters.
 View-up

YE View plane
coordinates

FIGURE 8-20
View-up and the view plane coordinates.

VRP

Z axis

Z axis
FIGURE 8-21
Changing the view reference point.

271
272 CHAPTER EIGHT

Changing the view plane normal will move the camera so as to photograph the
object from a different orientation, but the same part of the object is always shown at
the origin. (See Figure 8-22.)
Changing the view distance determines how far away*from the object the camera
is when it takes the picture. (See Figure 8-23.)
When changing view-up, we can imagine always pointing the camera at the
same object from the same direction, but twisting the camera in our hands so that the
picture, when developed, will turn out sideways or upside-down. (See Figure 8-24.)
These parameters allow the user to select how to view the object. Our system
must provide the user with a means of setting the parameters to the values which he de-
sires. The values are saved as global variables. The following algorithms provide an in-
terface between the user and the three-dimensional viewing transformation system.

8.14 Algorithm SET-VIEW-REFERENCE-POINT(X, Y, Z) For changing the

view reference point
Arguments X, Y, Z the new view reference point
Global XR, YR, ZR permanent storage for the reference point
BEGIN
XR < X;
Ra Y-
ZR <— Z;
RETURN;
END;

VPN
X axis
View plane
Zz axis

ane
X axis
Z axis
FIGURE 8-22
Changing the view plane normal.
 ae
V-dist | a

Z axis

ak X axis
V-dist { \\/

\/
Zz axis Y
FIGURE 8-23
Changing the view distance.

y axis

oe
ee

0
X axis

y axis

A View-up

X uxis

FIGURE 8-24
Changing the view-up direction.
273
274 CHAPTER EIGHT

8.15 Algorithm SET-VIEW-PLANE-NORMAL(DX, DY, DZ) For changing the view

plane normal
Arguments DX, DY, DZ the new view plane normal vector
Global DXN, DYN, DZN permanent storage for the View plane normal
Local D the-length of the user’s specification vector
Constant | ROUNDOFF some small number greater than any round-off error
BEGIN
D <— SQRT(DX f 2 + DY f 2 + DZ f 2);
IF D < ROUNDOFF THEN
RETURN ERROR ‘INVALID VIEW PLANE NORMAL’;
DXN <— DX/D;
DYN < DY/D,;
DZN < DZ/D;
RETURN;
END;

8.16 Algorithm SET-VIEW-DISTANCE(D) For changing distance between view

reference point and view plane
Argument _D the new distance
Global VIEW-DISTANCE the permanent storage for the view distance
BEGIN
VIEW-DISTANCE < D;
RETURN;
END;

8.17 Algorithm SET-VIEW-UP(DX, DY, DZ) For changing the direction which will be
vertical on the image
Global DXUP, DYUP, DZUP permanent storage for the view-up direction
Constant © ROUNDOFF some small number greater than any round-off error
BEGIN
IF |DX| + |DY| + |DZ| < ROUNDOFF THEN
RETURN ERROR ‘NO SET-VIEW-UP DIRECTION’;
DXUP < Dx;
DYUP < DY;
DZUP <— DZ;
RETURN;
END;

In the above algorithm we use an approximation to the length in order to avoid a

zero-length vector. It is easier and faster to sum absolute values than to calculate the
square root required for the true length (as we did in algorithm 8.15). The
approximation will be no smaller than the true length and will be larger by no more
than a factor of 1/3 . It therefore provides a good indication of the vector’s length and
will catch the zero cases.
We must also provide the user with a means for setting the projection. (See Fig-
ures 8-25 and 8-26.)
For a parallel projection, the user must specify the direction of the projection
lines. A perspective projection requires the location of the center of a projection point.
The values are saved in global variables. A flag is used to indicate whether a perspec-
Z axis

Z axis
FIGURE 8-25
Changing the parallel projection direction.

Z axis

Z axis |

FIGURE 8-26
Changing the center of perspective projection.
275
276 CHAPTER EIGHT

to whichever type
tive or a parallel projection is desired. The flag is set to correspond
of projection was last specified.
direction of
8.18 Algorithm SET-PARALLEL(DX, DY, DZ) For user input of the
parallel projection
Arguments DX, DY, DZ the new parallel projection vector
Global PERSPECTIVE-FLAG the perspective vs. parallel projection flag
DXP, DYP, DZP permanent storage for the direction of projection
Constant | ROUNDOFF some small number greater than any round-off error
BEGIN
IF |DX| + |DY| + |DZ| < ROUNDOFF THEN
RETURN ERROR ‘NO DIRECTION OF PROJECTION’;
PERSPECTIVE-FLAG < FALSE,
DXP < Dx;
DYP < DY;
DZP <— DZ;
RETURN;
END;

8.19 Algorithm SET-PERSPECTIVE(X, Y, Z) Indicate a perspective projection and

save the center of projection
Arguments X, Y, Z the new center of projection
Global XPCNTR, YPCNTR, ZPCNTR the permanent storage for the center
of projection
PERSPECTIVE-FLAG perspective vs. parallel projection flag
BEGIN
PERSPECTIVE-FLAG <- TRUE;
XPCNTR < X;
YPCNTR < Y;
ZPCNTR < Z;
RETURN;
END;

SPECIAL PROJECTIONS
The problem of rendering a two-dimensional view of a three-dimensional object exist-
ed long before the computer was used. One class of projection often used in drafting is
called an axonomertric projection. This is a parallel projection for.which the direction of
projection is perpendicular to the view plane. We can alter the direction of an
axonometric projection to get a different view of an object, provided we also change
the view plane normal to match the projection direction. Suppose that we are looking
at a cube that has edges parallel to the object coordinate axes. We might start with a
view looking straight at one of the faces so that the cube would appear to be a square.
(See Figure 8-27.)
If we change our direction of projection slightly to the side, then one of the side
faces will become visible, while the edges of the front face will shorten. If we raise our
angle of view, the top edges lengthen to make the top face visible, while the edges on
the side faces appear to shrink. (See Figure 8-28.)
 THREE DIMENSIONS 277

FIGURE 8-27
Looking straight at one of the faces of the cube.

There is a particular direction of projection for which all edges will appear short-
ened from their three-dimensional length by the same factor. This special direction is
called an isometric projection. (See Figure 8-29.)
An isometric projection of a cube will show a corner of the cube in the middle of
the image surrounded by three identical faces. From the symmetry of the situation, we
can see that the commands needed for an isometric projection of an object with sides
parallel to the axes are:

SET-PARALLEL(1, 1, 1);
SET-VIEW-PLANE-NORMAL(-1, -1, -1);

FIGURE 8-28
Changing the point of view shortens some edges and
lengthens others.
278 CHAPTER EIGHT

FIGURE 8-29
Isometric projection shortens all axes equally.

This projection will focus on the upper right-front corner of the object. The appropriate
commands for projections of the other seven corners are left to the reader as an exer-
cise.
If a viewing transformation is chosen such that edges parallel to only two of the
axes are equally shortened, then the projection is called dimetric. (See Figure 8-30.)
A trimetric projection is one in which none of the three edge directions is equally
shortened. There are no symmetries in the angles between the direction of projection
and the directions of the object edges. (See Figure 8-31.)
If the direction of parallel projection is not parallel to the view plane normal,
then we have what is called an oblique projection. We shall describe two special types
of oblique projections, called cavalier projections and cabinet projections.
Let us first discuss the cavalier projection. Let us assume that we are viewing an
object with edges parallel to the coordinate axes. The view plane will be parallel to the
front face (parallel to the xy plane). For a cavalier projection, the direction of projec-
tion is slanted so that points with positive z coordinates will be projected down and to
the left on the view plane. Points with negative z coordinates will be projected up and
to the right. The angle of the projected z axis (the ratio of up to right) can be whatever
we desire, but the distance the point shifts in the projected z direction must equal the
actual three-dimensional z distance from the view plane. The projection command
which will create a cavalier projection at angle A is

SET-PARALLEL(COS(A), SIN(A), 1).

Actually, any SET-PARALLEL command with arguments in the same ratio as those
above will work as well. (See Figure 8-32.) ‘
We stated that for a cavalier projection, the distance shiftedcalong the projected z
axis was equal to the actual z-axis distance. This restriction makes it easy to construct

FIGURE 8-30
Dimetric projection shortens two axes equally.
 THREE DIMENSIONS 279

‘FIGURE 8-31
Trimetric projection shortens all axes differently.

these drawings. However, the result is an object which appears elongated along the z
direction. An alternative which is still easy to construct with a scale and triangle is to
shift only half the actual z distance along the projected z axis. This is called a cabinet
projection. (See Figure 8-33.) It can be created by a SET-PARALLEL call with argu-
ments in the ratio of

SET-PARALLEL(COS(A), SIN(A), 2).

Perspective projections can be classified as one-point, two-point, or three-point.
These names refer to the number of vanishing points required for a construction of the
drawing. Our routines are sufficiently general to generate all three types of perspective
projection. A one-point perspective projection occurs when one of the faces of a rectan-
gular object is parallel to the view plane. (See Figure 8-34.)
A two-point perspective projection refers to the situation where one set of edges
runs parallel to the view plane, but none of the faces is parallel to it. (See Figure 8-35.)
A three-point perspective projection is the case where none of the edges is parallel
to the view plane. (See Figure 8-36.)

CONVERSION TO VIEW PLANE

COORDINATES
The user enters the description of the object in terms of the object coordinates. But our
particular point of view corresponds to expressing the object in the view plane coordi-
nates. The process of generating a particular view of the object is one of transforming
from one coordinate system to another. While this problem may appear complex, it is
actually one which we have already solved. The steps to be taken are the same as those

y axis

“A
A

gE:
it
FIGURE 8-32
x axis A cavalier projection.
280 CHAPTER EIGHT

y axis

“~*
A
ame AY:
FIGURE 8-33
x axis A cabinet projection.

which were used for a rotation about an arbitrary axis. We wish to perform aseries of
transformations which will change the object coordinates into the view plane coordi-
nates. The first step is a translation to move the origin to the correct position for the
view plane coordinate system. This is a shift first to the view reference point, and then
along the view plane normal by the VIEW-DISTANCE. After the origin is in place, we
align the z axis. In the rotation problem, we saw how to rotate a line onto the z axis.

FIGURE 8-34
A one-point perspective.
 ee FIGURE 8-35
Z axis A two-point perspective.

FIGURE 8-36
A three-point perspective.

281
 282 CHAPTER EIGHT

Here, we wish that line to be the object coordinate’s z axis, and we shall rotate it into
the view plane coordinate’s z axis (the view plane normal). This is done in two steps.
First, a rotation about the x axis places the line in the view plane coordinate’s xz plane.
Then a rotation about the y axis moves the z axis to its proper position. Now all that is
needed is to rotate about the z axis until the x and y axes are in their place in the view
plane coordinates. The entire transformation sequence is given by
TMATRIX = TR,R,R, (8.59)
where

1 0 0
0 1 0
= 0 0 1
-(XR + DXN * VIEW-DISTANCE) -(ZR + DZN * VIEW-DISTANCE)
-(YR + DYN * VIEW-DISTANCE)
(8.60)

1 0 0 0
R = 0 -DZN/V -DYN/V 0 4

x10 DYN/V -DZN/V 0 —

0 0 0 1

and

V = (DYN’ + DZN?)!? (8.62)

also

V- © -DXN 0
Ae el ono
7 IDXN: 08 UN ey (8.63)
oO 76 Ort
and

YUP-VP/RUP XUP-VP/RUP 0 0
-XUP-VP/RUP YUP-VP/RUP 0 0
Rz =>
0 0 0 (8.64)
0 0 Onl
where

[XUP-VP YUP-VP Z.1| = [DXUP DYUP DZUP 1] R,R, (8.65)

and

RUP = (XUP-VP? + YUP-VP?)!2 (8.66)

 THREE DIMENSIONS 283

What are these XUP-VP and YUP-VP variables? We are trying to finish up the transfor-
mation process by rotating the x and y axes into position. The correct position is ob-
tained when the y axis is aligned with the projection on the view plane of the view-up
vector. But the view-up vector was specified in the object coordinates. It is much
easier to work in the (as yet incomplete) view plane coordinate system. In the view
plane system, we can project the view-up vector onto the view plane by just ignoring
the z coordinate. The x and y values will be in the ratio of the sine and cosine of the
angle needed to correctly align the up direction. So what we are doing is performing
our partial transformation (R,R,) on the view-up direction. The translation part is not
needed because we are working on a vector (which has no position, only magnitude
and direction). The z coordinate is ignored to project onto the view plane, and the dis-
tance from the origin to the projected point is calculated. Dividing the x and y values
by this distance yields the sine and cosine of the needed rotation angle. The following
algorithm will create the view plane coordinate transformation.
8.20 Algorithm MAKE-VIEW-PLANE-TRANSFORMATION For making the view-
ing transformation
Global XR, YR, ZR the view reference point
DXN, DYN, DZN the view plane normal
DXUP, DYUP, DZUP the view-up direction
TMATRIX a 4 x 3 transformation matrix array
PERSPECTIVE-FLAG the perspective projection flag
VIEW-DISTANCE distance between view reference point and view plane
Local V, XUP-VP, YUP-VP, RUP for storage of partial results
Constant ROUNDOFF some small number greater than any round-off error
BEGIN
start with the identity matrix
NEW-TRANSFORM-3;
translate so that view plane center is new origin
TRANSLATE-3(— (XR + DXN * VIEW-DISTANCE),
—(YR + DYN * VIEW-DISTANCE), —(ZR + DZN * VIEW-DISTANCE));
rotate so that view plane normal is z axis
V<SQRT(DYN ¢ 2 + DZN 7 2);
IFV > ROUNDOFF THEN ROTATE-X-3(— DYN/V, — DZN/V);
ROTATE-Y-3( DXN, V);
determine the view-up direction in these new coordinates
XUP-VP <— DXUP * TMATRIX[I, 1] + DYUP * TMATRIX[2, 1] +
DZUP « TMATRIX{3, 1];
YUP-VP <— DXUP « TMATRIX(1, 2] + DYUP * TMATRIX[2, 2] +
DZUP * TMATRIX{3, 2];
determine rotation needed to make view-up vertical
RUP <— SQRT(XUP-VP 7 2 + YUP-VP 7 2);
IF RUP < ROUNDOFF THEN
RETURN ERROR ‘SET-VIEW-UP ALONG VIEW PLANE NORMAL;
ROTATE-Z-3(XUP-VP / RUP, YUP-VP / RUP);
IF PERSPECTIVE-FLAG THEN MAKE-PERSPECTIVE-TRANSFORMATION
ELSE MAKE-PARALLEL-TRANSFORMATION;
RETURN;
END;
284 CHAPTER EIGHT

The above algorithm also converts the parallel or perspective projection pa-
rameters from object coordinates to view plane coordinates. This is done by calling
either the MAKE-PARALLEL-TRANSFORMATION or the MAKE-PERSPECTIVE-
TRANSFORMATION routine. The MAKE-PERSPECTIVE-TRANSFORMATION
routine just applies the transformation matrix to the center of the projection point. An
error occurs if the center of projection is on the wrong side of the view plane, that is, if
the observer is on the same side of the screen as the object. The MAKE-PARALLEL-
TRANSFORMATION routine converts the direction of parallel projection to the view
plane coordinates. Since the direction is specified by a vector (no position), the transla-
tion portion of the transformation is omitted. An error occurs if the direction of projec-
tion turns out to be parallel to the view plane.

8.21 Algorithm MAKE-PERSPECTIVE-TRANSFORMATION Convert center of

projection to view plane coordinates
Global XPCNTR, YPCNTR, ZPCNTR the center of projection
XC, YC, ZC the center of projection in view plane coordinates
BEGIN
XC <— XPCNTR;
YC <— YPCNTR;
ZC <— ZPCNTR;
VIEW-PLANE-TRANSFORM(XC, YC, ZC);
IF ZC < 0 THEN
RETURN ERROR: “CENTER OF PROJECTION BEHIND VIEW PLANE’;
RETURN;
END;

8.22 Algorithm MAKE-PARALLEL-TRANSFORMATION Calculation of direction

of projection in view plane coordinates
Global TMATRIX a 4 X 3 coordinate transformation matrix array
DXP, DYP, DZP the parallel projection vector
VXP, VYP, VZP direction of projection in view plane coordinates
SXP, SYP the slopes of the projection relative to z direction
Constant ©ROUNDOFF some small number greater than any round-off error
BEGIN
VXP < DXP * TMATRIX[1, 1] + DYP * TMATRIX[2, the +
DZP * TMATRIX[3, 1);
VYP <— DXP * TMATRIX[I1, 2] + DYP * TMATRIX[2, 2] +
DZP * TMATRIX{3, 2];
VZP < DXP * TMATRIX[1, 3] + DYP « TMATRIX(2, 3] +
DZP * TMATRIX{[3, 3];
IF |VZP| < ROUNDOFF THEN
RETURN ERROR ‘PROJECTION PARALLEL VIEW PLANE’:
SXP <— VXP /VZP;
SYP — VYP/VZP;
RETURN;
END;
 THREE DIMENSIONS 285

CLIPPING IN THREE DIMENSIONS

In Chapter 6 we introduced the idea of a window, which served as a clipping boundary
in two-dimensional space. In three-dimensional space the concept can be extended to a
clipping volume or view volume. This is a three-dimensional region or box. Objects
within the view volume may be seen, while those outside are not displayed. Objects
crossing the boundary are cut, and only the portion within the view volume is shown.
A view volume may clip the front or back of an object as well as its sides. For exam-
ple, imagine the image of a building. The picture centers on the door of the building.
The size of the entrance increases as you seem to approach it. Passing through the
doorway, the outside walls disappear. The display now shows the entry hall. Clipping
may be used to remove the front wall of the building, which was hiding its interior.
(See Figure 8-37.)
The extension from two-dimensional to three-dimensional clipping is not a dif-
ficult one. The methods remain basically the same. The difference lies in the test to see
whether or not a point is inside the visible region. Instead of comparing the point
against a line, we now must compare the point against a plane. In general, any plane

y axis

Object

View
point

Zz axis
|
Clipping
plane

y axis

z axis
|
Front of object FIGURE 8-37
clipped away Moving a clipping plane through an object.
286 CHAPTER EIGHT

may be used as a boundary, and clipping regions can be arbitrary polyhedra; but there —
are two view volume shapes which are usually used because they are easy to work with. —
The type of view volume used depends on whether a parallel or a perspective projec- _
tion is to be employed. For a parallel projection, imagine planes that are in the direc-
tion of projection extending from the edges of the window. These planes form a rectan- _
gular tube in space. Front and back clipping planes can be added to section this tube
into a box. Objects within the box are visible, whereas those outside are clipped away.
(See Figure 8-38.) An application of front and back clipping is shown in Plate 2, where
the clipping planes are used to show a slice of a molecule. For a perspective projec-
tion, we picture rays from the center of projection passing through the window to form
a viewing pyramid. This pyramid can be truncated by the front and back clipping
planes to form the volume in which objects may be seen. (See Figure 8-39.)
The reason for choosing these regions becomes apparent when we consider what
happens when the objects are projected. After projection, objects within the view vol-
ume will lie within the window on the view plane. After projection, the clipping planes
run parallel to the z axis through the window boundaries, and their equations look the
same as the window boundary line equations. This means that if we project the objects
before clipping, then the left, right, top, and bottom clipping cases are essentially the
same as the window clipping we have already seen in Chapter 6.
We must provide a means to specify where the front and back clipping planes are
located. We shall provide algorithms to set these parameters in the same way that the
window boundary was specified. The planes will be positioned relative to the view ref-
erence point in the direction of the view plane normal. (See Figure 8-40.)
For both the parallel and perspective cases, we have a view volume bounded by
six clipping planes. The top, bottom, and side planes can be determined from the win-
dow and projection information, but additional routines are needed to allow the user to
specify the front and back clipping planes. This is just a matter of saving the position
specified by the user in some global variables. The following algorithm will save the
positions given in terms of the distance from the view reference point in the direction
of the view plane normal.

projection

View plane FIGURE 8-38

View volume for a parallel projection.
 THREE DIMENSIONS 287

View volume

Center
of
projection

View plane

FIGURE 8-39
View volume for a perspective projection.

8.23 Algorithm SET-VIEW-DEPTH(FRONT-DISTANCE, BACK-DISTANCE) User

routine to specify the position of front and back clipping planes
Arguments FRONT-DISTANCE, BACK-DISTANCE plane distance from the view
reference point along the view plane normal
Global FRONT-HOLD, BACK-HOLD storage for plane positions
BEGIN
IF FRONT-DISTANCE > BACK-DISTANCE THEN
“ RETURN ERROR ‘FRONT PLANE BEHIND BACK PLANE’;
FRONT-HOLD <— FRONT-DISTANCE;
BACK-HOLD < BACK-DISTANCE;
RETURN;
END;

We could, if we wished, omit the front and/or back clipping tests. All that is
needed is a flag which indicates whether to compare a point against the clipping plane

y axis
| |
BACK-DISTANCE
t
FRONT-DISTANCE

DISTANCE View
Z axis plane
VIEW- normal
REFERENCE-
POINT

View Front Back

plane plane plane

FIGURE 8-40
Front and back clipping plane specification.
288 CHAPTER EIGHT

or to pass it to the next routine without checking. With very little effort, we can give
the user the ability to turn the front and back clipping on or off. Routines which will
set clipping flags for this purpose are given below. .

8.24 Algorithm SET-FRONT-PLANE-CLIPPING(ON-OFF) User routine to set the

front clipping flag
Argument ON-OFF the user’s clipping flag setting
Global FRONT-FLAG-HOLD the front clipping flag set by the user
BEGIN
FRONT-FLAG-HOLD < ON-OFF;
RETURN;
END;

8.25 Algorithm SET-BACK-PLANE-CLIPPING(ON-OFF) User routine to set the

back clipping flag
Argument ON-OFF the user’s clipping flag setting
Global BACK-FLAG-HOLD the back clipping flag set by the user
BEGIN
BACK-FLAG-HOLD < ON-OFF;
RETURN;
END;

We need to decide whether to perform the z-plane clipping before or after projec-
tion. Clipping can be done at either point, provided that the Z/(Z- — Z) transformation
is used on the clipping plane position if we are clipping after a perspective projection.
The argument for clipping in z after the projection is that CLIP-FRONT and CLIP-
BACK algorithms can simply be incorporated into the clipping sequence. The reason
why clipping before projection is desirable is that perspective projection requires that
objects lie behind the center of projection. Now if we wish to construct programs
where the center of projection (the eye of the viewer) moves among objects and
perhaps even through them, then we would like to remove from consideration all ob-
jects which lie in front of the center of projection before that projection is carried out.
That is, we want to employ front clipping to remove the objects that the viewer has
passed and then to project the remaining objects which might be seen. (See Figure
8-41.) We shall present algorithms for the simpler case of clipping after projection and
leave clipping before projection as a programming problem. ;
We shall use the MAKE-Z-CLIP-PLANES algorithm to capture and save the
user’s front and back clipping specification (which can be changed at any time). This
will give fixed clipping characteristics for the life of a display-file segment, as is done
in NEW-VIEW-2 for the window clipping planes.

8.26 Algorithm MAKE-Z-CLIP-PLANES Establish the front and back clipping planes
Global FRONT-HOLD, BACK-HOLD storage for plane positions
FRONT-Z position of the front clipping plane
BACK-Z position of the back clipping plane
FRONT-FLAG-HOLD the front clipping flag set by the user
BACK-FLAG-HOLD the back clipping flag set by the user
 THREE DIMENSIONS 289

Viewpoint

View Plane Clipping Plane

Viewpoint

View Plane Clipping Plane

FIGURE 8-41
Front clipping before projection can remove objects that should not be projected.

FRONT-FLAG the front clipping flag

BACK-FLAG the back clipping flag
VIEW-DISTANCE the permanent storage for the view distance
PERSPECTIVE-FLAG the perspective projection flag
XC, YC, ZC the center of projection in view plane coordinates
BEGIN
FRONT-FLAG < FRONT-FLAG-HOLD;
BACK-FLAG < BACK-FLAG-HOLD;
FRONT-Z <— VIEW-DISTANCE — FRONT-HOLD;
BACK-Z <— VIEW-DISTANCE — BACK-HOLD;
IF PERSPECTIVE-FLAG THEN
BEGIN
FRONT-Z < FRONT-Z / (ZC — FRONT-Z);
BACK-Z <— BACK-Z / (ZC — BACK-Z);
END;
RETURN;
END;
290 CHAPTER EIGHT

CLIPPING PLANES
The clipping problem becomes one of deciding on which side of a plane a point lies.
This is similar to the problem of deciding on which side of a line a point lies in a
plane. Remember that the equation of a plane has the form
Ax + By + 'Cz-+ D= 0 (8.67)
Suppose we plug the coordinate values of a point (x,, y;, Z;) into this equation. Then if
the point is on the plane, the equation will be satisfied.
Ax, + By, +Cz,;+D=0 (8.68)
But if the point is not on the plane, the result will not be zero. It will be positive if the
point is on one side of the plane
Ax, + By, + Cz, + D>0 (8.69)
and negative if the point is on the other side.
Ax, + By, + Cz, + D<0 (8.70)
We can therefore tell if a point is within the clipping boundary by checking the sign of
the expression obtained from the equation of the plane.
Let’s consider what the equations describing our actual clipping planes will be.
The front and back clipping planes are particularly simple. We shall take them to be
parallel to the view plane, so in the view plane coordinate system they are given by
z = FRONT-Z and z= BACK-Z (8.71)
where FRONT-Z and BACK-Z are constants giving the positions of these planes on the
z axis. (See Figure 8-42.)
For the point (x;, y;, Z;) to be visible, it must be behind or on the front plane and
in front of or on the back plane. The tests should then be

Z, = FRONT-Z and Z, 2 BACK-Z (8.72)

y axis

Visible part

Z axis

X axis FIGURE 8-42

Front and back clipping.
 THREE DIMENSIONS 291

FIGURE 8-43
The clipping process.

Using these tests, we may write algorithms similar to those of Chapter 6 for clipping
against back and front planes. These algorithms may be included with upgraded ver-
sions of the two-dimensional algorithms to give us three-dimensional clipping. Each
visible point passes from one algorithm to the next until it has been checked against all
clipping planes. (See Figure 8-43.)

8.27 Algorithm CLIP-BACK(OP, X, Y, Z) Routine for clipping against the back

boundary
Arguments OP, X, Y, Z a display-file instruction
Global BACK-Z position of the back clipping plane
BACK-FLAG the back clipping flag
XS, YS, ZS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF BACK-FLAG THEN
BEGIN
IF PFLAG AND NEEDFIRST[5] THEN
BEGIN
FIRSTOP[5] < OP;
FIRSTX[5] <— X;
FIRST Y[5] —Y;
FIRSTZ[5] <— Z
NEEDFIRST[5] <— FALSE;
END
ELSE
IF Z < BACK-Z AND ZS[5] < BACK-Z THEN
292 CHAPTER EIGHT

CLIP-FRONT(1, (X — XS[5]) * (BACK-Z — Z)/(Z — ZS[5]) + X,

(Y — YS[5]) * (BACK-Z — Z)/(Z — ZS[5]) + Y,
BACK-Z)
ELSE es
IFZ < BACK-Z AND ZS[5] > BACK-Z THEN
IF OP > 0 THEN
CLIP-FRONT(OP, (X — XS[5]) * (BACK-Z — Z)/(Z — ZS[5])
+ X,
(Y — YS[5]) * (BACK-Z — Z)/(Z — ZS[5])
+ Y, BACK-Z)
ELSE
CLIP-FRONT(1, (X — XS[5]) * (BACK-Z — Z)/(Z — ZS[S5])
+X,
(Y — YS[5]) * (BACK-Z — Z)/(Z — ZS[5])
+ Y, BACK-Z);
XS[5] <— X;
YS[5] — Y;
ZS[5] — Z;
IF Z = BACK-Z AND CLOSING # 5 THEN CLIP-FRONT(OP, X, Y, Z);
END
ELSE CLIP-FRONT(OP, X, Y, Z);
RETURN;
END;

8.28 Algorithm CLIP-FRONT(OP, X, Y, Z) Routine for clipping against the front

boundary
Arguments OP, X, Y, Z a display-file instruction
Global FRONT-Z position of the front clipping plane
FRONT-FLAG the front clipping flag
XS, YS, ZS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF FRONT-FLAG THEN
BEGIN
IF PFLAG AND NEEDFIRST[6] THEN
BEGIN es
FIRSTOP[6] < OP;
FIRSTX[6] <— X;
FIRSTY[6] — Y;
FIRSTZ[6] <— Z;
NEEDFIRST[6] <— FALSE;
END
ELSE
IF Z = FRONT-Z AND ZS[6] > FRONT-Z THEN
CLIP-LEFT(1, (X — XS[6]) * (FRONT-Z — Z)/(Z — ZS[6]) + X,
(Y — YS[6]) * (FRONT-Z — Z)/(Z — ZS[6]) + Y,
FRONT-Z)
 THREE DIMENSIONS 293

ELSE
IF Z = FRONT-Z and ZS[6] < FRONT-Z THEN
IF OP > 0 THEN
CLIP-LEFT(OP, (X — XS[6]) * (FRONT-Z — Z)/(Z — ZS[6])
a8 2s
(Y — YS[6]) * (FRONT-Z — Z) /(Z — ZS[6})
+ Y, FRONT-Z)
ELSE
CLIP-LEFT(1, WXH, (X — XS[6]) « (FRONT-Z — Z) /
(Z — ZS[6]) + X;
(Y — YS[6]) * (FRONT-Z — Z)/
(Z — ZS[6]) + Y, FRONT-Z);
XS[6] <— X;
YS[6] — Y;
ZS[6] <— Z;
IF Z = FRONT-Z AND CLOSING # 6 THEN CLIP-LEFT(OP, X, Y, Z);
END;
ELSE CLIP-LEFT(OP, X, Y, Z);
RETURN;
END;

The following routines are extensions of those presented in Chapter 6. In many

cases all that is needed is to include a z coordinate along with x and y. In the four
routines which actually do the clipping, the z coordinate of any intersection with the
clipping plane must be calculated, as well as the x or the y coordinate. We are main-
taining the z-coordinate information in anticipation of its use in the hidden-surface
routines of the next chapter.

8.29 Algorithm CLIP-LEFT(OP, X, Y, Z). (An extension of algorithm 6.6 to three

dimensions) Routine for clipping against the left boundary
Arguments OP, X, Y, Z a display-file instruction
Global WXL window left boundary
XS, YS, ZS arrays containing the last point drawn
3
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[1] THEN
BEGIN
FIRSTOP[1] <— OP;
FIRSTX[1] <— X;
FIRSTY[1] — Y;
FIRSTZ[1] <— Z;
NEEDFIRST[1] < FALSE;
END
Case of drawing from outside in
ELSE
IF X = WXL AND XS[1] < WXL THEN
294 CHAPTER EIGHT

CLIP-RIGHT(1, WXL, (Y — YS[1]) * (WXL — X)/(K — XS[1]) + Y,

(Z — ZS[1]) * (WXL — X)/(K — XS[1]) + Z)
Case of drawing from inside out .
ELSE ;
IF X = WXL AND XS[1] > WXL THEN
IF OP > 0 THEN
CLIP-RIGHT(OP, WXL, (Y — YS[1]) * (WXL — X)/(X — XS[1])
cays
(Z — ZS[1]) * (WXL — X)/(% — XS[1))

ELSE
+D)
CLIP-RIGHT(1, WXL, (Y — YS[1]) * (WXL — X)/(XK — XS[1))
ar eY,
(Z — ZS[1}) * (WXL — X)/(X — XS[1))
am 4)
Remember point to serve as one of the endpoints of next line segment
XS[1] — X;
YSULESye
ZS[llicsZs
Case of point inside |
IF X = WXL AND CLOSING # 1 THEN CLIP-RIGHT(OP, X, Y, Z); |
RETURN; ‘0 |
END;

8.30 Algorithm CLIP-RIGHT(OP, X, Y, Z) (An extension of algorithm 6.7 to three

dimensions) Routine for clipping against the right boundary
Arguments OP, X, Y, Z a display-file instruction
Global WXH window right boundary |
XS, YS, ZS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[2] THEN
BEGIN
FIRSTOP[2] < OP;
FIRSTX[2] <— X;
FIRSTY[2] < Y: ce
FIRSTZ[2] < Z;
NEEDFIRST[2] < FALSE; :
END
ELSE
IF X < WXH AND XS[2] > WXH THEN
CLIP-BOTTOM(1, WXH, (Y — YS[2]) * (WXH — X)/(X — XS[2]) + Y,
(Z_— ZS[2]) * (WXH — X)/(X — XS[2]) + Z)
ELSE
IF X = WXH and XS[2] < WXH THEN
IF OP > 0 THEN
CLIP-BOTTOM(OP, WXH, (Y — YS[2]) * (WXH — X) /(K — XS[2])
crane
 THREE DIMENSIONS 295

(Z — ZS[2]) * (WXH — X)/(XK — XS[2])

+ Z)
ELSE eae
CLIP-BOTTOM(1,
WXH, (Y — YS[2]) * (WXH — X)/(X — XS[2])
ye
(Z = ZS} = (WKH = X)/(X - XS[2))
ae 74)
XS[2] <— X; ieent
YS[2] <Y;
ZS[2] <—Z;
IF X = WXH AND CLOSING # 2 THEN CLIP-BOTTOMC(OP, X, Y, Z);
RETURN;
END;

8.31 Algorithm CLIP-BOTTOM(OP, X, Y, Z) (An extension of algorithm 6.8 to three

dimensions) Routine for clipping against the lower boundary
Arguments OP, X, Y, Z a display-file instruction
Global WYL window lower boundary
XS, YS, ZS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[3] THEN
BEGIN
FIRSTOP[3] <— OP;
FIRSTX[3] <— X;
FIRSTY[3] <—Y;
FIRSTZ[3] <— Z;
NEEDFIRST[3] < FALSE;
END
ELSE
IFY = WYL AND YS[3] < WYL THEN
e CLIP-TOP(1, (K — XS[3]) * (WYL — Y)/(Y — YS[3]) + X, WYL,
(Ze ESS) ce OW Nd) XG 9 (5) ete Z)
ELSE
IF Y = WYL AND YS[3] > WYL THEN
IF OP > 0 THEN
CLIP-TOP(OP, (XK — XS[3]) * (WYL — Y)/(Y — YS[3]) + X, WYL,
(Z — ZS[3)) = (WYL. = Y/Y — YSI3)) + Z)
ELSE
CLIP-TOP(1, (X — XS[3]) * (WYL — Y)/(Y — YS[3]) + X, WYL,
(Z — ZS[3]) * (WYL — Y)/(Y — YS[3]) + Z);
XS[3] <— X;
YS[3] — Y;
ZS[3] <— Z;
IFY = WYL AND CLOSING # 3 THEN CLIP-TOP(OP, X, Y, Z);
RETURN;
END;
296 CHAPTER EIGHT

8.32 Algorithm CLIP-TOP(OP, X, Y, Z) (An extension of algorithm 6.9 to three

dimensions) Routine for clipping against the upper boundary
Arguments OP, X, Y, Z a display-file instruction
.
Global WYH window upper boundary
XS, YS, ZS arrays containing the last point drawn
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
BEGIN
IF PFLAG AND NEEDFIRST[4] THEN
BEGIN
FIRSTOP[4] <— OP;
FIRSTX[4] <— X;
FIRSTY[4] <— Y;
FIRSTZ[4] < Z,;
NEEDFIRST[4] < FALSE;
END
ELSE
IF Y= WYH AND YS[4] > WYH THEN
SAVE-CLIPPED-POINT(1, (X — XS[4]) * (WYH — Y)/
(Y — YS[4]) + X, WYH,
(Z — ZS[4]) * (WYH — Y)/(Y — YS[4]) + Z);
ELSE
IFY = WYH AND YS[4] < WYH THEN
IF OP > 0 THEN
SAVE-CLIPPED-POINT(OP, (X — XS[4]) * (WYH — Y)/
(Y — YS[4]) + X, WYH,
(Z — ZS[4]) * (WYH — Y)/
(Y — YS[4]) + Z)
ELSE
SAVE-CLIPPED-POINT(1, (X — XS[4]) * (WYH — Y)/
(Y — YS[4]) + X, WYH,
(Z — ZS[4]) * (WYH — Y)/
(Y = YS[4)) + Z);
XS[4] — X;
YS[4] <Y;
ZS[4] <—Z;
IF Y = WYH AND CLOSING # 4 THEN SAVE-CLIPPED-PQINT(OP, XEYeZ):
RETURN; 7 a
END;

8.33 Algorithm SAVE-CLIPPED-POINT(OP, X, Y, Z) (An extension of

algorithm
6.10 to three dimensions) Saves clipped polygons in T-buffer and sends lines
and charac-
ters to the display file
Arguments OP, X, Y, Z a display-file instruction
Global COUNT-OUT a counter of number of sides on clipped polygon
PFLAG indicates if a polygon is being clipped
BEGIN
IF PFLAG THEN
BEGIN
 THREE DIMENSIONS 297

COUNT-OUT < COUNT-OUT + 1:

PUT-IN-T(OP, X, Y, Z, COUNT-OUT);
END i.
ELSE VIEWING-TRANSFORM(OP, X, Y);
RETURN;
END;

8.34 Algorithm PUT-IN-T(OP, X, Y, Z, INDEX) (An extension of algorithm 6.11 to

three dimensions)
Arguments OP, X, Y, Z the instruction to be stored
INDEX the position at which to store it
Global IT, XT, YT, ZT arrays for temporary storage of polygon sides
BEGIN
IT[INDEX] < OP;
XT[INDEX] <— X;
YT[INDEX] <— Y;
ZT[INDEX] <— Z;
RETURN;
END;

8.35 Algorithm CLIP-POLYGON-EDGE(OP, X, Y, Z) (An extension of algorithm

6.12 to three dimensions) Close and enter a clipped polygon into the display file
Arguments OP, X, Y, Za display-file instruction
Global PFLAG indicates that a polygon is being drawn
5 COUNT-IN the number of sides remaining to be processed
COUNT-OUT the number of sides to be entered in the display file
IT, XT, YT, temporary storage arrays for a polygon
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
FRONT-FLAG, BACK-FLAG indicate whether front and back clipping is
done
Local I for stepping through the polygon sides
BEGIN
COUNT-IN <— COUNT-IN — 1;
CLIP-BACK(OP, X, Y, Z);
IF COUNT-IN 4 0 THEN RETURN;
close the clipped polygon
CLOSING < 5;
IF BACK-FLAG AND NOT NEEDFIRST[S] THEN
CLIP-BACK(FIRSTOP[5], FIRSTX[5], FIRSTY[5], FIRSTZ[5]);
CLOSING < 6;
IF FRONT-FLAG AND NOT NEEDFIRST[6] THEN
FIRSTX[6],
CLIP-FRONT(FIR STOP[6] , FIRSTZ[6]);
FIRSTY([6],
CLOSING <I; | |Ly MOT Atay s
IF NOT NEEDFIRST[1] THEN
CLIP-LEFT(FIRSTOP[1], FIRSTX[1], FIRSTY[1], FIRSTZ[1));
CLOSING < 2;
IF NOT NEEDFIRST[2] THEN
CLIP-RIGHT(FIRSTOP[2], FIRSTX[2], FIRSTY[2], FIRSTZ[2]);
298 CHAPTER EIGHT

CLOSING < 3;
IF NOT NEEDFIRST([3] THEN
CLIP-BOTTOM(FIRSTOP([3], FIRSTX[3], FIRSTY(3], FIRSTZ[3]);
CLOSING < 4; .
IF NOT NEEDFIRST[4] THEN
CLIP-TOP(FIRSTOP[4], FIRSTX[4], FIRST Y[4], FIRSTZ[4]);
CLOSING < 0;

PFLAG < FALSE;

IF COUNT-OUT < 3 THEN RETURN;
enter the polygon into the display file
VIEWING-TRANSFORM(COUNT-OUT, XT[COUNT-OUT], YT[COUNT-OUT));
FOR I = 1 TO COUNT-OUT DO VIEWING-TRANSFORM(T{I], XT[I], YT[I]);
RETURN:
END;

8.36 Algorithm CLIP(OP, X, Y, Z) (An extension of algorithm 6.13 to three dimen-

sions) Top-level clipping routine a
Arguments OP, X, Y, Z the instruction being clipped
Global PFLAG indicates that a polygon is being processed
COUNT-IN number of polygon sides still to be input
COUNT-OUT number of clipped polygon sides stored
XS, YS, ZS arrays for saving the last point drawn
Local I for initializing the four clipping routines
BEGIN
IF PFLAG THEN CLIP-POLYGON-EDGE(OP., X, Y, Z)
ELSE IF OP > 2 THEN +
BEGIN
PFLAG <— TRUE;
COUNT-IN < OP;
COUNT-OUT < 0;
FORI = 1TO6DO
BEGIN
XS[I] — X;
YS(] <Y;
ZS[I] <— Z;
END;
END
ELSE CLIP-BACK(OP, X, Y, Z); ‘
RETURN;
END;

THE 3D VIEWING TRANSFORMATION

We finally have all of the tools which are needed to process the user’s three-dimen-
sional drawing. We assume that the user has specified the desired viewing parameters,
and that the MAKE-VIEW-PLANE-TRANSFORMATION routine has been used to
create a view plane coordinate transformation. The user now calls a three-dimensional
 THREE DIMENSIONS 299

drawing command such as LINE-ABS-3(X, Y, Z). This command updates the pen po-
sition and calls the DISPLAY-FILE-ENTER routine which transforms and projects the
position onto the view plane before entering it into the display file. The steps are to
first multiply by the view plane transformation matrix to convert the point to view
plane coordinates. This in effect changes the viewpoint to that of our synthetic camera.
Second, we perform a projection to place the two-dimensional image on the ‘‘film.’’
Finally, we save the image in the display file by means of our clipping routine. Doing
the clipping after the view plane transformation makes it seem as if our clipping win-
dow is attached to the view plane. It governs the size of the ‘‘film’’ in the synthetic
camera. (See Figure 8-44.)
An algorithm to multiply a point by the view plane transformation matrix and
thereby convert it from object to view plane coordinates is given next.

8.37 Algorithm VIEW-PLANE-TRANSFORM(X, Y, Z) Transforms a point into the

view plane coordinate system
Arguments X, Y, Z point to be transformed, also for return of result
Global TMATRIX a 4 X 3 transformation matrix array
Local T three-element array to hold results until calculation finished
I index for stepping through the TMATRIX columns
BEGIN
FOR I = 1TO3 DO
T{I] — X « TMATRIX[1, I] + Y * TMATRIX[2, I] +
- Zx* TMATRIX(3, I] + TMATRIX(4, I);
X<—T[]];
Y¥<—T[2);
Z <—T[3];
RETURN;
END;

FIGURE 8-44
The window is attached to the view plane.
300 CHAPTER EIGHT

The modified DISPLAY-FILE-ENTER routine is as follows:

8.38 Algorithm DISPLAY-FILE-ENTER(OP) (Revisiog of algorithm 6.14) Routine

to enter an instruction into the display file
Argument OP opcode of instruction to be entered
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
PERSPECTIVE-FLAG perspective vs. parallel projection flag
Local X, Y, Z hold the point that is transformed
BEGIN
IF OP < 1 AND OP > —32 THEN PUT-POINT(OP, 0, 0)
ELSE
BEGIN
X <— DF-PEN-X;
Y <— DF-PEN-Y;
Z <— DF-PEN-Z;
VIEW-PLANE-TRANSFORM(X, Y, Z);
IF PERSPECTIVE-FLAG THEN PERSPECTIVE-TRANSFORM(X, Y, Z)
ELSE PARALLEL-TRANSFORM(X, Y, Z);
CLIP(OP, X, Y, Z);
END;
RETURN;
END;

To finish up, we will extend the CREATE-SEGMENT routine to call an al-

gorithm named NEW-VIEW-3 that forms a new viewing transformation and establishes
clipping parameters. This means that a new viewing transformation can be established
for each display-file segment. The three-dimensional viewing parameters are therefore
established in the same manner as the window specification. The user may set new
values at any time, but the new values will not go into effect until a display-file seg-
ment is created. Furthermore, a particular viewing specification will remain in effect
throughout the segment.

8.39 Algorithm NEW-VIEW-3 Create a new overall viewing transformation

BEGIN
MAKE-VIEW-PLANE-TRANSFORMATION;
NEW-VIEW-2;
MAKE-Z-CLIP-PLANES;
RETURN; ‘
END;

8.40 Algorithm CREATE-SEGMENT(SEGMENT-NAME) (Modification of

algorithm 7.17) User routine to create a named segment
Argument SEGMENT-NAME the segment name
Global NOW-OPEN the segment currently open
FREE the index of the next free display-file cell
SEGMENT-START, SEGMENT-SIZE, VISIBILITY, ANGLE, SCALE-X,
SCALE-Y, TRANSLATE-X, TRANSLATE-Y,
DETECTABLE the segment-table arrays
Constant | NUMBER-OF-SEGMENTS size of the segment table
 THREE DIMENSIONS 301

BEGIN
IF NOW-OPEN > 0 THEN RETURN ERROR ‘SEGMENT STILL OPEN’;
IF SEGMENT-NAME < 1 OR SEGMENT-NAME > NUMBER-OF-SEGMENTS
THEN
RETURN ERROR ‘INVALID SEGMENT NAME’;
IF SEGMENT-SIZE[SEGMENT-NAME] > 0 THEN
RETURN ERROR ‘SEGMENT ALREADY EXISTS’;
NEW-VIEW-3
SEGMENT-START[SEGMENT-NAME] < FREE;
SEGMENT-SIZE[SEGMENT-NAME] < 0;
VISIBILITY [SEGMENT-NAME] < VISIBILITY [0];
ANGLE[SEGMENT-NAME] < ANGLE(0];
SCALE-X[SEGMENT-NAME] < SCALE-X[0];
SCALE-Y [SEGMENT-NAME] < SCALE-Y(0];
TRANSLATE-X[SEGMENT-NAME] <— TRANSLATE-X[(0];
TRANSLATE-Y[SEGMENT-NAME] <— TRANSLATE-Y [0];
DETECTABLE[SEGMENT-NAME] < DETECTABLE(0];
NOW-OPEN < SEGMENT-NAME;
RETURN;
END;

We also need an initialization routine to establish default viewing parameters.

The default parameters we shall choose are a view reference point at the origin, the
view plane in the object coordinate xy plane, the view-up along the object coordinate’s
y direction, and a parallel projection in the z direction. Front and back clipping will ini-
tially be off.
8.41 Algorithm INITIALIZE-8 Initialization of global data
Local I for initialization of the clipping routines
BEGIN
initialize
INITIALIZE-7;
SET-VIEW-REFERENCE-POINT(0, 0, 0);
SET-VIEW-PLANE-NORMALA(O, 0, — 1);
_ SET-VIEW-DISTANCE(O);
SET-VIEW-UP(O, 1, 0);
SET-PARALLEL(O, 0, 1);
SET-FRONT-PLANE-CLIPPING(FALSE);
SET-BACK-PLANE-CLIPPING(FALSE);
SET-VIEW-DEPTH(0, 0);
NEW-VIEW-3;
FOR I = 1 to6 DO
BEGIN
NEEDFIRST[I] <— FALSE;
XS[I] — 0;
YS[{I] — 0;
ZS[I] — 0;
END;
RETURN;
END;
302 CHAPTER EIGHT

AN APPLICATION
One application relying heavily upon three-dimensional graphics is an airplane flight
simulator. A flight simulator can be used as part of a pilet’s training. The simulator
may look like the real cockpit, only the windshield is replaced by a computer-gener-
ated image of the world. This image alters under the pilot’s actions in the same manner
as his view of the world would change if he were actually flying. In the simulator, the
pilot may practice, and even err, without endangering anything but his pride. (See Fig-
ure 8-45.)
Let us consider how our three-dimensional graphics system might be used in a
flight simulator program. The first thing to be done is to construct a model of the world
over which the pilot is to fly. Buildings, runways, fields, lakes, and other landscape
features may be constructed using our three-dimensional LINE and POLYGON primi-
tives. Windowing allows us to use real-world dimensions, such as meters. The con-
struction may be somewhat tedious, but it is straightforward. Let us assume that it has
been done, that we have a procedure named BUILD-WORLD which contains the com-
mands for drawing the model landscape. We must still project this landscape onto the
display screen. We wish to do it in such a manner as to produce the view which the
pilot should see from his airplane. We shall consider the view plane to be the cockpit
windshield. The view reference point will be attached to the airplane, so as it moves,
the windshield (view plane) moves with it. A VIEW-DISTANCE value of zero will do
nicely.
It may be convenient to think of the view reference point as centered on the
windshield. We can accomplish this with our SET-WINDOW command.

SET-WINDOW(
—0.5, 0.5, —0.5, 0.5);

The orientation of the plane corresponds to the orientation of its windshield, which is
set by the SET-VIEW-PLANE-NORMAL command. The pilot may bank or roll the
plane. This would change its view-up direction. And finally, we must indicate the pro-
jection, which should be a perspective view centered on the pilot’s eye. This is behind
the windshield, say about 0.5 meters away. (See Figure 8-46.)
Let us outline two routines, one which is called whenever the plane changes its
position, and another to be called when the pilot alters the orientation of the plane. To
update the plane’s position, we change the view reference point and the center of pro-
jection. :
“

B
FIGURE 8-45
Airplane flight simulation.
 THREE DIMENSIONS 303

Bank angle

Center of projection
View plane

View reference
point
Window

FIGURE 8-46
Parameters for flight simulation.

BEGIN
SET-VIEW-REFERENCE-POINT(EAST-WEST, ALTITUDE, NORTH-SOUTH);
SET-PERSPECTIVE(EAST-WEST — 0.5 x EW-DIRECTION,
ALTITUDE — 0.5 « AL-DIRECTION, NORTH-SOUTH
— 0.5 * NS-DIRECTION);
CREATE-SEGMENT(2);
“SET-VISIBILITY(2, FALSE);
BUILD-WORLD;
CLOSE-SEGMENT;
SET-VISIBILITY(1, FALSE);
SET-VISIBILITY(2,TRUE);
DELETE-SEGMENT(1);
RENAME-SEGMENT(2, 1);
RETURN;
END;

Here we have used (EAST-WEST, ALTITUDE, NORTH-SOUTH) as the coordi-

nates of the plane’s position. The vector [EW-DIRECTION AL-DIRECTION NS-
DIRECTION] gives the plane’s orientation and is assumed to be normalized. In the
SET-PERSPECTIVE command, 0.5 is used as the distance of the pilot from the wind-
shield. It is multiplied by the plane’s direction and subtracted from the plane’s position
to give the pilot’s position. The visibility property and segment renaming are used to
ensure that the old image of the world is maintained until the new one is ready for dis-
play.
A routine to update the plane’s orientation might look something like the follow-
ing:

BEGIN
SET-VIEW-PLANE-NORMAL(EW-DIRECTION, AL-DIRECTION, NS-DIREC-
TION);
SET-VIEW-UP(SIN(BANK-ANGLE) * NS-DIRECTION,
304 CHAPTER EIGHT

COS(BANK-ANGLE) + (EW-DIRECTION { 2 + NS-DIRECTION f 2) f 0.5,

— SIN(BANK-ANGLE) * EW-DIRECTION);
RETURN;
END; .

In the above, BANK-ANGLE is how much the plane is banked to the left. A
value of zero means level flight, and a negative value means banking to the right. The
arithmetic which occurs within the call to SET-VIEW-UP finds a vector in the world
coordinates which, if attached to the plane, would correspond to the vertical direction
in level flight.
A program for flight simulation might also take advantage of three-dimensional
clipping. Such a program should not display what is behind the airplane. We should ex-
tend the program so that it clips any object which lies behind the windshield (the view
plane). We can do this with our front clipping plane. If the pilot has unlimited visibil-
ity, then the back clipping plane should not be used. If, on the other hand, we wish to
simulate the limiting effects of bad weather on visibility, an approach might be to sim-
ply clip away all objects that exceed the range of vision.
Instructions to set up the front and back clipping might look as follows:

SET-VIEW-DEPTH(0, VISIBILITY-DISTANCE);
SET-FRONT-PLANE-CLIPPING(TRUE);
SET-BACK-PLANE-CLIPPING(BAD-WEATHER);

In the above, VISIBILITY-DISTANCE is the distance which the pilot is able to see.
BAD-WEATHER is TRUE if the pilot’s visibility is limited, and FALSE otherwise.

FURTHER READING
The mathematics of transformations and projections is further discussed in [ROG76].
An excellent discussion of homogeneous coordinates and transformations is available
in [AHU68]. A mathematical discussion of rotations and reflection matrix properties is
given in [FIL84]. There is a fine discussion of the homogeneous form of points, lines,
and planes in [BLI77]. Homogeneous coordinates and the relationship between pro-
jected and Euclidean space are considered in [RIE81]. Homogeneous coordinates and
clipping are considered in [BLI78]. A formal description of transformations and clip-
ping ina hierarchical picture structure is described in [MAL78]. A discussion of pro-
jections and how they are specified is given in [CAR78]. A discussion of how to use
our viewing system is presented in [BER78]. The viewing process is described in
[ROG83]. In [MIC80] there is a critical evaluation of the CORE viewing parameters
which we have used, and an alternative scheme for describing the view is presented.
Perspective projections help to give some realism to three-dimensional objects but still
only give a two-dimensional image. Work has been done on devising display hardware
that will create three-dimensional images by presenting a different view to each eye.
A j

survey of these techniques is found in [LAN82]. We have shown how to describe

three-
dimensional objects by their two-dimensional polygon surfaces, but they may
also be
described by three-dimensional structures. This is called solid modeling and is
often
used for computer-aided design. Three-dimensional image representation is also
useful
 THREE DIMENSIONS 305

for medical applications such as computerized tomography. Some of these techniques

are described in [AGG81], [BAD78], [HER82], [JAC80], [MEA82], [REQ80],
[SRI81], and [WOO77]. A discussion of flight simulation may be found in [SCH81].

[AGG81] Aggarwal, J. K., Davis, L. S., Martin, W. N., Roach, J. W., ““Survey: Representation Methods
for Three-Dimensional Objects,’’ Progress in Pattern Recognition, vol. 1, pp. 377-391 (1981).
[AHU68] Ahuja, D. V., Coons, S. A., ‘‘Geometry for Construction and Display,’’ JBM Systems Journal,
vol. 7, nos. 3 & 4, pp. 188-205 (1968).
[BAD78] Badler, N., Bajcsy, R., ‘‘Three-Dimensional Representations for Computer Graphics and Com-
puter Vision,’’ Computer Graphics, vol. 12, no. 3, pp. 153-160 (1978).
[BER78] Bergeron, R. D., Bono, P. R., Foley, J. D., ‘‘Graphics Programming Using the CORE System,”’
ACM Computing Surveys, vol. 10, no. 4, pp. 389-394 (1978).
[BLI77] Blinn, J. F, ‘‘A Homogeneous Formulation for Lines in 3-Space,’’ Computer Graphics, vol. 11, no.
2, pp. 237-241 (1977).
[BLI78] Blinn, J. F, Newell, M. E., ‘‘Clipping Using Homogeneous Coordinates,’’ Computer Graphics,
vol. 12, no. 3, pp. 245-251 (1978).
[CAR78] Carlbom, I., Paciorek, J., ‘‘Planar Geometric Projections and Viewing Transformations,’’ ACM
Computing Surveys, vol. 10, no. 4, pp. 465-502 (1978).
[FIL84] Fillmore, J. P., ‘‘A Note on Rotation Matrices,’’ JEEE Computer Graphics and Applications, vol. 4,
no. 2, pp. 30-33 (1984).
([HER82] Herman, G. T., Reynolds, R. A., Udupa, J. K., ‘‘Computer Techniques for the Representation of
Three-Dimensional Data on a Two-Dimensional Display,’’ Proceedings of the SPIE, vol. 367, pp. 3-
14 (1982).
[JAC80] Jackins, C. L., Tanimoto, S. L., “‘Oct-Trees and Their Use in Representing Three-Dimensional Ob-
jects,’’ Computer Graphics and Image Processing, vol. 14, no. 3, pp. 249-270 (1980).
[LAN82] Land, B., ‘‘Stereoscopic Displays,’’ Proceedings of SPIE, vol. 367, pp. 20-32 (1982).
[MAL78] Mallgren, W. R., Shaw, A. C., “‘Graphical Transformations and Hierarchic Picture Structures,”’
Computer Graphics and Image Processing, vol. 8, no. 3, pp. 237-258 (1978).
[MEA82] Meagher, D., ‘‘Geometric Modeling Using Octree Encoding,’’ Computer Graphics and Image Pro-
cessing, vol. 19, no. 2, pp. 129-147 (1982).
[MIC80] Michener, J. C., Carlbom, I. B., ‘‘Natural and Efficient Viewing Parameters,’’ Computer
Graphics, vol. 14, no. 3, pp. 238-245 (1980).
[REQ80] Requicha, A. A. G., ‘‘Representations for Rigid Solids: Theory, Methods, and Systems,’’ Comput-
ing Surveys, vol. 12, no. 4, pp. 437-464 (1980).
(RIE81] Riesenfeld, R. EF, ‘‘Homogeneous Coordinates and Projective Planes in Computer Graphics,”’
«IEEE Computer Graphics and Applications, vol. 1, no. 1, pp. 50-55 (1981).
[ROG76] Rogers, D. EF, Adams, J., A., Mathematical Elements for Computer Graphics, McGraw-Hill, New
York (1976).
[ROG83] Rogers, G., ‘‘Viewing,’’ Computers & Graphics, vol. 7, no. 1, pp. 83-85 (1985).
[SCH81] Schachter, B. J., ‘“Computer Image Generation for Flight Simulation,’’ JEEE Computer Graphics
and Applications, vol. 1, no. 10, pp. 29-68 (1981).
{SRI81] Srihari, S. N., ‘‘Representation of Three-Dimensional Digital Images,’’ Computing Surveys, vol.
13, no. 4, pp. 399-424 (1981).
[WOO77] Woo, T. C., ‘‘Progress in Shape Modeling,’’ Computer, vol. 10, no. 12, pp. 40-46 (1977).

EXERCISES
8-1 Write the parametric form for the line passing through the following points:
a) (0, 0, 0) and (1, 2, 3)
b) (1, 5, 4) and (2, 2, 3)
c) (8, 4, 1) and (5, 2, — 1)
306 CHAPTER EIGHT

8-2 Give an equation for the following planes:

a) Containing the point (0, 0, 0) and normal to vector [0, 0, 1]
b) Containing the point (0, 0, 0) and normal to vector [0, 0, — 1]
c) Containing the point (0, 0, 0) and normal to vector [1, 1, 0]
d) Containing the point (1, 1, 2) and normal to vector [4, 2, 3]
e) Containing the points (1, 0, 0), (0, 1, 0), and (0, 0, 1)
f) Containing the points (0, 0, 0), (1, 0, 0), and (0, 1, 1)
g) Containing the points (0, 0, 0), (1, 2, 3), and (2, 3, 1)
h) Containing the points (— 1, 0, 1), (0, 1, 0), and (—1, 1, —2)
8-3 Write the three-dimensional homogeneous transformation matrix for each of the following
transformations:
a) Shift 0.5 in x, 0 in y, and —0.2 in z.
b) Scale z to be half as large.
c) Scale x andy to be three times as large.
d) Rotate by 7/4 about the x axis (y into z).
e) Rotate by 7/3 about the y axis (x into z).
f) Rotate by 7 about the line passing through the points (0, 0, 0) and (1, 0, 1).
8-4 Suppose we wish to rotate about the origin so that a point (x, y, Z) is moved to the positive z
axis. Prove that the following four methods all result in the same transformation matrix.
a) Rotate about the x axis to place the point in the xz plane. Then rotate about the y axis
to place the point on the z axis.
b) Rotate about the y axis to place the point in the yz plane. Then rotate about the x axis
to place the point on the z axis.
c) Rotate about the z axis to place the point in the xz plane. Then rotate about the y axis
to place the point on the z axis. ,
d) Rotate about the z axis to place the point in the yz plane. Then rotate about the x axis
to place the point on the z axis.
8-5 Given the object

y axis

(0.5, 0.8, -0.2)

(0.2, 0.5, -0.8)

Z axis

X axis

(0.8, 0.2,-0.2) \ (0.8, 0.2, -0.8)

(0.2, 0.2, -0.2)

show how it would appear on the screen for each of the following viewing parameter specifica-
tions.
 THREE DIMENSIONS 307

a) Default settings except

SET-VIEW-REFERENCE-POINT(0.2, 0.2, —0.2)
b) Default settings except
SET-PARALLEL(0.2, 0, 1)
c) Default settings except
SET-PARALLEL(— 0.2, 0, 1)
d) Default settings except
SET-PARALLEL(O, —0.2, 1)
e) Default settings SET-PARALLEL(O, 0.2, 1)
except
f) Default settings except
SET-VIEW-PLANE-NORMAL(— 0.5, 0, — 1)
g) Default settings SET-VIEW-PLANE-NORMAL(0, 0.5, — 1)
except
h) Default settings SET-PARALLEL(0.5, 0, 1)
except
SET-VIEW-PLANE-NORMAL(— 0.5, 0, — 1)
i) Default settings except SET-PARALLEL(O, 0.5, 1)
SET-VIEW-PLANE-NORMALA(O, — 0.5, — 1)
j) Default settings except SET-PARALLEL(0.5, 0.5, 1)
SET-VIEW-PLANE-NORMAL(— 0.5, —0.5, —1)
k) Default settings except SET-PARALLEL(0.5, 0, 1)
SET-VIEW-PLANE-NORMAL(-— 0.5, 0, — 1)
SET-VIEW-REFERENCE-POINT(0.2, 0.2, —0.2)
1) Default settings except SET-PARALLEL(1, 0, 0)
SET-VIEW-PLANE-NORMAL( — 1,0, 0)
SET-VIEW-REFERENCE-POINT(0.2, 0.2, = 02)
m) Default settings except SET-PARALLEL(0.2, 0, 1)
SET-VIEW-DISTANCE(0.2)
n) Default settings except SET-PARALLEL(0.2, 0, 1)
SET-VIEW-DISTANCE( — 0.2)
0) Default settings except SET-VIEW-PLANE-NORMAL(-— 1, 0, 0)
SET-PARALLEL(1, 0, 0)
SET-VIEW-REFERENCE-POINT(0.8, 0.2, —(0°2)
p) Default settings except SET-VIEW-UP(1, 1, 0)
q) Default settings except SET-PARALLEL(O, 1, 0)
SET-VIEW-PLANE-NORMAL(O, — 1, 0)
SET-VIEW-REFERENCE-POINT(0.2, 0.2, =)
SET-VIEW-UP(O, 0, — 1)
r) Default settings except SET-VIEW-UP(— 1, 0, 0)
SET-VIEW-REFERENCE-POINT(0.8, 0.2, =e}
SET-VIEW-PLANE-NORMAL(O, — 1, 0)
ie SET-PARALLEL(O, 1, 0)
8-6 Consider the figure in Exercise 8-5. With the default viewing parameter except for
SET-PERSPECTIVE(0.5, 0.5, 1)
SET-DEPTH-CLIPPING-PLANES(0.3, 0.7)
sketch what would be shown for
a) SET-FRONT-CLIPPING(TRUE)
SET-BACK-CLIPPING(FALSE)
b) SET-FRONT-CLIPPING(FALSE)
SET-BACK-CLIPPING(TRUE)
c) SET-FRONT-CLIPPING(TRUE)
SET-BACK-CLIPPING(TRUE)
d) SET-VIEW-REFERENCE-POINT(0, 0, =) 7)
SET-FRONT-CLIPPING(TRUE)
SET-BACK-CLIPPING(TRUE)
308 CHAPTER EIGHT

8-7 a) Prove that the perspective projection description expressed in Equation 8.46 is
equivalent to that for Equation 8.55.
b) Prove that the perspective projection description expressed in Equation 8.55 is
equivalent to that for Equation 8.56. Pp
8-8 Prove that the perspective projection of a line segment is equal to the line segment between
the perspective projection of the endpoints.
8-9 Develop three-dimensional clipping algorithms which may be applied before projection.
8-10 Develop a general three-dimensional clipping algorithm for clipping against an arbitrary
plane. The clipping plane should be specified by the arguments to your algorithm as well as by
the line to be clipped. Recursive or repeated calls to your algorithm should allow clipping to
view volumes that are arbitrary convex polyhedra.

PROGRAMMING PROBLEMS
8-1 Implement algorithms 8.1 through 8.41 and extend your graphics system to three
dimensions.
8-2 Test your graphics system by entering and viewing a three-dimensional house. Show the
house in both parallel and perspective projection. Test the action of each of the viewing
parameters by making small changes and observing the result.
8-3 Test the three-dimensional clipping algorithms by constructing a three-dimensional object
model, such as a house, and in separate views for each, clip using the front, back, top, bottom,
and side clipping planes. Try for both parallel and perspective projections.
8-4 Write the routines
INQUIRE-VIEW-REFERENCE-POINT(X, Y, Z)
INQUIRE-VIEW-PLANE-NORMAL(DX, DY, DZ)
INQUIRE-VIEW-DISTANCE(D)
INQUIRE-VIEW-UP(DX, DY, DZ)
INQUIRE-PARALLEL(ON-OFF, DX, DY, DZ)
INQUIRE-PERSPECTIVE(ON-OFF, X, Y, Z)
INQUIRE-DEPTH-CLIPPING-PLANES(FRONT, BACK)
INQUIRE-FRONT-CLIPPING(ON-OFF)
INQUIRE-BACK-CLIPPING(ON-OFF)
which return to the user the last settings for the viewing parameters.
8-5 a) Show a cube in isometric, dimetric, and trimetric projection.
b) Show a cube in cavalier and in cabinet projection.
c) Show a cube in one-point, two-point, and three-point perspective projection.
8-6 Construct a program which gives successive views of a three-dimensional object generated
by procedure FOO. The views should give the effect of walking around the object. Assume that
the object is contained within the cube with corners (0, 0, 0) and (1, 1, — 1),
8-7 Write a program which given the function

Y = FUN(X, Z)
will show the indicated surface by constructing lines of constant z for z = 0.2,
0.4, 0.6, 0.8 by
connecting the values at x = 0.2, 0.4, 0.6, 0.8. Select a viewing transformation which
best dis-
plays the surface. Try this for FUN(X, Z) = SIN(X) * SIN(Z), or FUN(X, Z) = X « Z, or
PUNKS Z)o= 1X f 2 Z to.
 THREE DIMENSIONS 309

a
a
8-8’ Display some solid object in perspective projection. Turn on the front clipping and, in suc-
cessive views, move the clipping plane through the object by changing the view reference point
and the center of projection. (Caution: If your routines clip after projection, be careful that the
center of projection stays in front of the object.)
8-9 Write a program to move the viewer through a world of objects by setting the view reference
point to successive positions along a path. Set the view plane normal to the vector between suc-
cessive view reference point positions.
8-10 For a truly three-dimensional image of a scene we must present separate stereo views of
the scene to each eye (corresponding to the two separate viewpoints). A number of methods have
been tried in order to present each eye with its own picture, including systems of mirrors and
lenses, shutters, polarized filters, and color filters. Write a graphics program which will produce
the two stereo views required for three-dimensional display. (Hint: Consider the viewing parame-
ter settings which will generate the scene as seen by each eye.)
310 CHAPTER EIGHT

8-11 Write a procedure which takes a polygon as an argument and enters into the graphics sys-
tem a number of instances of the polygon rotated about the z axis. Using the procedure create a
cylinder from a rectangle, a sphere from a circle.
8-12 Write a three-dimensional interactive point plotting program. Use three valuators to set the
X, y, and z positions. Echo the position by displaying three intersecting cross hairs with tic
marks.
*§-13 Construct a font of characters which can be positioned in three-dimensional space.
Generalize the TEXT procedure to allow placement ofa string of characters in any orientation on
any three-dimensional plane. Show some text written on the top, sides, and bottom of a box.
Display it using both parallel and perspective projections.
*§8-14 Write an interactive program which repeatedly draws some three-dimensional object, al-
lowing the user to select from a menu and alter the three-dimensional viewing and clipping pa-
rameters.
*8-15 Revise your graphics system so that the front and back clipping is carried out prior to pro-
jection.
*8-16 Write procedures for clipping against a view volume bounded by six arbitrary planes.
 NINE
HIDDEN
SURFACES
AND
LINES

INTRODUCTION
In the last chapter we described how to obtain different views of a scene. We de-
veloped perspective projections which made constructed objects look more realistic.
But for a realistic scene, we should only draw those lines and polygons which could
actually be seen, not those which would be hidden by other objects. What is hidden
and what is visible depend upon the point of view. As seen from the front, the front of
a building is visible, while the back of the structure is hidden; but as seen from the
rear, this situation is reversed. We cannot see the contents of the building from outside
because they are hidden by the building’s walls, but from a point of view inside the
building, some of the contents should be displayed. So far, we have learned how to
model and project three-dimensional objects, but all parts of the objects are always dis-
played. This gives our drawings a transparent quality. Such figures are called wire-
frame drawings because they look as if they are wire outlines of the supposedly solid
objects. Complex objects can easily turn into a confusing clutter of line segments. It
may be difficult to judge which lines belong to the front of the object and which to the
back. The removal of hidden portions of objects is essential to producing realistic-look-
ing images. (See for example Plates 4 through 15.) In this chapter we consider the
problem of removing those lines which would normally be hidden by part of the ob-
ject. If we can assign this task to the machine, then the user will be free to construct
the entire model (front, back, inside, and outside) and still be able to see it as it will ac-
tually appear. This problem is not nearly as easy as it might seem at first. What nature

311
312 = CHAPTER NINE

does with ease (and a lot of parallel processing) we must do through extensive compu-
tation. There exist many solutions to the hidden surface and line problem. Many ap-
proaches have been tried; perhaps the best solution has yet to be found. We shall dis-
cuss several approaches, but only one solution will be implemented. The first section
considers back-face detection and removal. This is sufficient for single convex objects.
The second section considers the removal of hidden surfaces by means of the painter's
algorithm.

BACK-FACE REMOVAL
Hidden-line removal can be a costly process, and so it behooves us to apply easy tests
to simplify the problem as much as possible before undertaking a thorough analysis.
There isa simple test which we can perform which will eliminate most of the faces
which cannot be seen. This test identifies surfaces which face away from the viewer.
They are the surfaces which make up the back of the object. They cannot be visible
because the bulk of the object is in the way. This does not completely solve the hidden-
surface problem because we can still have the front face of an object obscured by a
second object or by another part of itself. But the test can remove roughly half of the
surfaces from consideration and thus simplify the problem.
We begin our discussion by noting that we shall only consider polygons. Lines
cannot obscure anything, and although they might be obscured, they are usually found
only as edges of surfaces on an object. Because of this, polygons suffice for most
drawings. Now a polygon has two surfaces, a front and a back, just as a piece of paper
does. We might picture our polygons with one side painted light and the other painted
dark. But given the arrays of vertex coordinates which represent polygons, how can we
tell which face is which? Easy! We shall say that when we are looking at the light sur-
face, the polygon will appear to be drawn with counterclockwise pen motions. If we
move our point of view to the other side, so that the dark surface may be seen, the pen
drawing the polygon will appear to move clockwise. (See Figure 9-1.)

FIGURE 9-1
Drawing direction changes with the side viewed.
 HIDDEN SURFACES AND LINES 313

Now alittle mathematics will give us the direction that the polygon surface is
facing (the normal to the plane of the polygon). We shall describe two methods of find-
ing the vector normal to the plane of the polygon. The first method uses an operation
that is called the vector cross product. The cross product of two vectors is a third vector
with the length equal to the product of the lengths of the two vectors times the sine of
the angle between them and, most important to us, a direction perpendicular to the
plane containing the two vectors. (See Figure 9-2.)
The formulas for a three-dimensional cross product in a right-handed coordinate
system are as follows. For

ee Se Pe PD Ee <Q, 0, 0.) (9.1)

we have

Rx ae P.O, ee PQ;

Ry = POs oan FQ, (9.2)

R, = POs, = PQ,

Now two sides of a polygon describe two vectors in the plane of the polygon, so the
cross product of two polygon sides forms a vector pointing out from the polygon face.
Will this vector point out from the dark face or the light face? That depends on whether
the two sides form a convex or concave angle.
Let’s assume that we are dealing with two adjacent sides which do not lie in the
same line and which meet in a convex vertex, that is, a vertex where two sides meet to
form a convex angle. The vector cross product will yield a vector which points out of
the light face. (See Figure 9-3.)
The problem with this first approach to finding the vector normal to the polygon
is that some searching and checking are required to find a vertex at a convex corner of
noncollinear sides. The second method (which we shall use) is suggested by Newell
and described in [SUT74]. The calculation is as follows: If the n vertices of the poly-
gon are (x;, y;, Z;), then form the sums over all vertices

R-

FIGURE 9-2
R=PXQ P The vector cross product.
314 = CHAPTER NINE

FIGURE 9-3
The cross product point out of the light face at a convex vertex.

n
a= 2% OG: ee)
1=

s ll F 2 = Z;)(Xj oe X;) (9.3)

1=

n
° ll ame Cam XY; is yj)
i=1

where ifi = n, thenj = 1; otherwise, } = i+ 1.

The result [a bc] is a vector normal to the polygon. Each of these sums gives
twice the area of the projections of the polygon on a plane. That is, if we project the
polygon along the x direction to the yz plane, then the area of that projected polygon is
a/2. The values b/2 and c/2 are the areas of the projections on the xz and xy planes, re-
spectively. (The proof of this is left as an exercise.) So a, b, and c describe the projec-
tion of the polygon in the x, y, and z directions. But this is directly related to the direc-
tion of the plane. The amount of area projected along the z direction, for example, is
proportional to the z component of the polygon’s normal vector. So[a bc] is a nor-
mal vector to the plane of the polygon. The virtue of this method is that there are no
special cases to check; we simply compute the sums.
Now, suppose that we make the rule that all solid objects are to be constructed
out of polygons in such a way that only the light surfaces are opem to the air; the dark
faces meet the material inside the object. This means that when we look at an object
face from the outside, it will appear to be drawn counterclockwise. (See Figure 9-4.)
If a polygon is visible, the light surface should face toward us and the dark sur-
face should face away from us. Since a cross product or sum can be formed which
gives the direction of the light face, this vector should point toward us. So if the nor-
mal vector points toward the viewer, the face is visible (a front face), otherwise, the
face is hidden (a back face) and should be removed.
How can we tell whether or not a vector points toward the viewer? To do this, we
examine the z component of the normal vector. If the z component is positive, then the
polygon faces toward the viewer; if negative, it faces away.
 HIDDEN SURFACES AND LINES 315

FIGURE 9-4
All exterior faces are colored light (drawn counterclock-
wise

This is a special case of the general problem of comparing two vectors. If we

have two vectors (say R and S) and wish to compare their directions, we use the vector
dot product.
a =R-S = [R||S| cos 6 (9.4)
As we saw in Chapter 8, the dot product gives the product of the lengths of the two
vectors times the cosine of the angle between them. This cosine factor is important to
us because if the vectors are in the same direction (0 = 0 < 7/2), then the cosine is
positive and the overall dot product is positive; but if the directions are opposing
( 1/2 < @ = 77), then the cosine and the overall dot product are negative. (See Figure
9-5.)
_ The formula for computing a dot product is as follows:
a— RS

[ReARy RSs ySd (9.5)

= R,S, + RySy + RS,

For the back-face check, one vector is the normal to the polygon and the other is the
depth direction [0 0 1]. So the test for a back face is then a check on the sign of the
z component of the normal vector. (See Figure 9-6.)
Where should all of this checking be done? Certainly it should be done after the
transformation to view plane coordinates. It should follow projection (the projection
can affect whether the polygon image is drawn clockwise or counterclockwise). We
should also wait until after clipping has been performed so that we won't bother with
polygons outside of the window. There is a point in the CLIP-POLYGON-EDGE

cos @ > 0 6 <0

cos

FIGURE 9-5 ’
and negative if they
Cosine of the angle.is positive if the vectors are pointing somewhat in the same direction
point away from each other.
316 CHAPTER NINE

Result of
cross product

FIGURE 9-6
The left plane would be visible from the left because
its cross product and projection vectors point in the
same direction (positive dot product). The right plane
Projection vectors would be a back face.

routine where the clipped polygon is sitting in a temporary storage buffer waiting to be
placed into the display file. We test to be sure that the number of sides is at least 3 be-
fore the polygon is saved. At this point we can also perform our back-face check, re-
quiring the polygon to also be a front face before it is saved.

BACK-FACE ALGORITHMS
Now let us detail the algorithms which will remove back faces. To begin with, we may
not always want hidden lines removed, so we should provide the user with the means
of turning the process on or off.

9.1 Algorithm SET-HIDDEN-LINE-REMOVAL(ON-OFF) User routine to set the

hidden-line removal indicator
Argument ON-OFF user specification for removing hidden lines
Global HIDDEN hidden-line removal flag.
BEGIN
HIDDEN < ON-OFF;
RETURN;
END;

The following is the modified CLIP-POLYGON-EDGE routine, which now in-

cludes a check for back faces when the HIDDEN flag is true.

9.2 Algorithm CLIP-POLYGON-EDGE(OP, X, Y, Z) (Revision of algorithm 8.35)

Close and enter a clipped polygon into the display file
Arguments OP, X, Y, Z a display-file instruction
Global PFLAG indicates that a polygon is being drawn
COUNT-IN the number of sides remaining to be processed
COUNT-OUT the number of sides to be entered in the display file
IT, XT, YT temporary storage arrays for a polygon
NEEDFIRST array of indicators for saving the first command
FIRSTOP, FIRSTX, FIRSTY, FIRSTZ arrays for saving the first command
CLOSING indicates the stage in polygon
 HIDDEN SURFACES AND LINES 317

FRONT-FLAG, BACK-FLAG indicate whether front and back clipping is

done
HIDDEN flag for hidden-line removal
Local I for stepping through the polygon sides
BEGIN
COUNT-IN <— COUNT-IN — 1;
CLIP-BACK(OP, X, Y, Z);
IF COUNT-IN # 0 THEN RETURN;
close the clipped polygon
CLOSING < 5;
IF BACK-FLAG AND NOT NEEDFIRST[5] THEN
CLIP-BACK(FIRSTOP[5], FIRSTX[5], FIRSTY[5], FIRSTZ[5]);
CLOSING < 6;
IF FRONT-FLAG AND NOT NEEDFIRST[6] THEN
CLIP-FRONT(FIRSTOP[6], FIRSTX[6], FIRSTY[6], FIRSTZ[6];
CLOSING < 1;
IF NOT NEEEDFIRST[1] THEN
CLIP-LEFT(FIRSTOP{1], FIRSTX[1], FIRSTY[1], FIRSTZ[1]);
CLOSING < 2;
IF NOT NEEDFIRST[2] THEN
CLIP-RIGHT(FIRSTOP[2], FIRSTX(2], FIRSTY([2], FIRSTZ[2]);
CLOSING < 3;
IF NOT NEEDFIRST[3] THEN
- CLIP-BOTTOM(FIRSTOP{[3], FIRSTX[3], FIRSTY[3], FIRSTZ[3]);
CLOSING < 4;
IF NOT NEEDFIRST[4] THEN
CLIP-TOP(FIRSTOP[4], FIRSTX[4], FIRSTY[4], FIRSTZ[4]);
CLOSING < 0;
PFLAG < FALSE;
IF COUNT-OUT < 3 THEN RETURN;
enter the polygon into the display file
IF HIDDEN THEN CALL BACK-FACE-CHECK(COUNT-OUT);
BES eeols abhor its KH Ad) BUSH VT ae
BEGIN
VIEWING-TRANSFORM(COUNT-OUT, XT[COUNT-OUT],
YT[COUNT-OUT]);
FOR I = 1 TO COUNT-OUT DO
VIEWING-TRANSFORM(IT[]], XT[], YT[I));
END;
RETURN;
END:

The algorithm above uses the BACK-FACE-CHECK function to actually decide

whether the polygon is a back face and enters it into the display file. It computes the z
component of the vector normal to the polygon according to Equation 9.3. It then
checks the sign and enters the polygon only if the z component is positive. Thus, only
front faces are entered.
318 =cHAPTER NINE

9.3 Algorithm BACK-FACE-CHECK(POLYSIZE) Filters out polygon drawn

clockwise
Argument POLYSIZE the number of sides on the polygon
Global XT, YT, ZT T-buffer array storage of the vertex points
Local C z component ofa vector for the normal to the plane of the polygon
I, J for stepping through the vertices
BEGIN
C<—0;
FOR I = 1 TO POLYSIZE DO
BEGIN
IFT = POLYSIZE THEN J < 1
ESE Wicd ls
CHC (Cth — XT)) = (YT) YE);
END;
IF C = 0 THEN RETURN;
VIEWING-TRANSFORM(POLYSIZE, XT[POLYSIZE], YT[POLYSIZE]);
FOR I = 1 TO POLYSIZE DO VIEWING-TRANSFORM(T[I], XT[], YT[I));
RETURN;
END;

We have added a new global flag, so we should include it in the initializations.

We shall make the default hidden-line removal setting FALSE, so that hidden lines will
not be removed unless such action is explicitly requested by the user.

9.4 Algorithm INITIALIZE-9A [Initialization routine

BEGIN
INITIALIZE-8;
SET-HiIDDEN-LINE-REMOVAL(FALSE);
RETURN;
END;

We have just seen how to remove many of the lines which would be hidden by an
object’s bulk. The method will suffice for single convex objects, but may be in-
adequate when several objects or concave surfaces are involved. (See Figure 9-7.) We
now survey some of the techniques used to solve the full hidden-psurace and hidden-
line problems.

FIGURE 9-7
Hidden lines among front faces.
 HIDDEN SURFACES AND LINES 319

Z BUFFERS
Consider raster displays backed by a frame buffer. These displays can show surfaces
(filled polygons) as well as lines. If we are using such a display, another way to state
the hidden-surface problem is that we want to arrange the frame buffer so that the color
displayed at any pixel is that of the surface closest to the viewer for that point. To do
this, we must somehow compare all of the surfaces which are projected onto the pixel
and decide which one can be seen. We must sort the polygons according to their posi-
tion in space. This notion of geometrically sorting the surfaces is central to hidden sur-
face and line removal.
One simple approach to this problem relies on a device called a Z buffer. The Z
buffer was described by Catmull [CAT74]. It is a large array with an entry for each
pixel on the display (like a frame buffer). The Z buffer is used to save the z coordinate
values. It helps us to sort out the polygons by keeping track of the z position of the sur-
faces being displayed. When the frame buffer is cleared, the Z-buffer elements are all
set to a very large negative value (a value which is beyond anything which will be im-
aged). The initial value may be thought of as the z position of the background. Poly-
gons will be entered one by one into the frame buffer by the display-file interpreter,
using scan conversion algorithms such as those discussed in Chapter 3. Suppose that
the algorithm which turns on each pixel of the polygon knows the projected z position
of the point being displayed. It could then compare the z position of the polygon point
with the Z-buffer value and decide if the new surface is in front of or behind the cur-
rent contents of the frame buffer. If the new surface has a z value greater than the Z-
buffer value, then it lies in front; its intensity value is entered into the frame buffer, and
its z value is entered into the Z buffer. If the z value of the new surface is less than the
value in the Z buffer, then it lies behind some polygon which was previously entered.
The new surface will be hidden and should not be imaged. No frame buffer or Z-buffer
entries will be made. The comparison is carried out on a pixel-by-pixel basis, and we
must be able to find the z position of the projected point for every pixel of the polygon.
If we were to modify our system to use a Z buffer for hidden-surface removal,
we might extend our display file to keep the z coordinates of the polygon vertices.
From the z coordinates of the vertices, we can find the z coordinate of any interior
point. We have seen in Chapter 3 (Algorithm 3.15 UPDATE-X-VALUES) how to incre-
mentally find the x values for the polygon edges as we step through the scan lines. The
same technique can be used to find the z values along the edges at each step. All we
need is the starting z value and the change in z for each step in y for the polygon
edges. So for each scan line, then, we still generate pairs of x coordinates which indi-
cate where to fill; but in addition, we know a z value for each of the fill-span end-
points. By reapplying this linear interpolation technique, this time using the span’s
change in z for each step in the x direction, the z values at each pixel can be calcu-
lated. THE FILLIN algorithm would have to obtain the endpoint z coordinates as argu-
ments, perform the interpolation in x, and compare the resulting z values to the Z-buf-
fer values to decide whether to change the frame buffer intensity.
A Z buffer can be expensive. It requires a lot of memory (one entry for each
pixel), and each entry must have a sufficient number of bits to distinguish the possible
320 CHAPTER NINE

z values. It can also be time-consuming in that a decision must be made for every pixel
instead of for the entire polygon. However, it is a very simple method, simple enough
to implement in hardware to overcome the speed problems, And the time required to
process a scene is proportional to the number of objects in the scene. (Some other tech-
niques which compare every polygon against every other polygon require time propor-
tional to the square of the number of polygons.) This, together with the continuing
drop in the cost of memory, makes this method (or its extensions) an increasingly
popular approach to the hidden-surface problem.

SCAN-LINE ALGORITHMS
Actually, if we are willing to collect the polygons and process them together, then a
full-screen Z buffer is not needed; a scan-line Z buffer will suffice and requires much
less memory. The idea was suggested by Carpenter [CAR76]. The reason a large
amount of memory is needed for a Z buffer is that we process each polygon indepen-
dently. As each polygon is rasterized, we must in effect be able to remember the depth
of each of its pixels so that they may be compared against later polygons. A polygon
may be as large as the screen, so we need a full-screen Z buffer. But we can reduce the
memory requirements by processing all of the polygons together on a scan-line by
scan-line basis. This in effect is repeatedly doing a Z-buffer hidden-surface removal
for a screen that is only one pixel high (a single scan line). The Z buffer only needs to
hold one scan line’s worth of depth information. When the scan line is done, we save
the result, reinitialize the Z buffer, move to the next scan line, and do the next scan-
line Z-buffer sort. The key here is that all polygons are processed together. Instead of
considering all scan lines for a polygon before moving on to the next polygon, we con-
sider all polygons for a scan line before moving on to the next scan line. This can be
done using algorithms much like those of Chapter 3, only instead of entering the edges
of a single polygon into the edge list, we enter the sides of all the polygons. We must
also be careful when pairing edges for filling that we pair edges belonging to the same
polygon.
A variation of the scan-line approach does not require a Z buffer at all. Where a
scan line cuts a polygon, a line segment or span is described. It is these spans which
are being sorted for the scan line’s hidden-surface removal. To order them, we need
only determine the depth of the spans at a few points. The interesting points are the
span endpoints (and intersection points if polygon faces can penetrate one another). At
the x position where a span begins, we could compare its depth with that of the other
spans active at that point to decide which is closest to the viewer. Likewise, when a
span ends, we could determine which of the remaining spans active at that point
should be shown. But often we can get by with even less work. Usually each scan line
looks much like its neighbors and we can use this to simplify the calculation. If faces
do not interpenetrate, and the order in x of the span ends does not change from one
scan line to the next, then the depth ordering is the same for the two scan lines and no
new sorting in z is needed. Depth reordering is needed only when the sweep through
the scan lines encounters a new polygon, passes a polygon, or finds a change in the
order of span endpoints.
 HIDDEN SURFACES AND LINES 32]

THE PAINTER’S ALGORITHM

There is a property of frame buffers used by Newell [NEW72], which has become
known as the painter's algorithm. The algorithm gets its name from the manner in which
an oil painting is created. The artist begins with the background. He can, if he wishes,
fill the entire canvas with the background scene. The artist then paints the foreground
objects. There is no need to erase portions of the background; the artist simply paints
on top of them. The new paint covers the old so that only the newest layer of paint is
visible. A frame buffer has this same property. We enter a filled polygon into the frame
buffer by changing the proper pixels to values corresponding to the polygon’s interior
style. If we then enter a second polygon ‘‘on top of’’ the first, some of those same
pixels will be changed to correspond to the second polygon’s interior style. Wherever
the second polygon lies, the first polygon’s pixel settings have been forgotten. The sec-
ond polygon has ‘‘covered up’’ the first. The painter’s algorithm tells us to enter first
those polygons which are farthest from the viewer (the background) and enter last the
objects closest to the viewer (the foreground). Hidden surfaces can be covered up by
choosing the correct order to draw them and taking advantage of the properties of
frame buffers. (See Figure 9-8.)
The painter’s algorithm is a simple idea, but let’s look at what is involved in its
implementation. First of all, we cannot process each polygon independently, as we did

So?

FIGURE 9-8
The most recent filled polygon overwrites previous pixel values.
322 CHAPTER NINE

in the back-face check or the full-screen Z buffer. We must compare each polygon with
all of the rest to see which is in front of which. We must, in effect, sort the polygons to
determine a priority for their display. The sorting will determine the order in which
they will be entered into the display file (the order in which they will be drawn). For ef-
ficiency, we try to limit the number of comparisons and we try to make the comparison
process fast.

COMPARISON TECHNIQUES
There are several techniques for determining the relevancy and relative position of two
polygons. Not all tests may be used with all hidden-surface algorithms, and some of
the tests are not always conclusive. Sometimes we can use simple tests for many of the
cases and resort to more costly tests only when all else fails.
One technique that is often useful is called the minimax test or boxing test. We may
not need to know the order of every polygon relative to every other polygon; it may
suffice to know just the relative orderings of those polygons that overlap. So a test that
can quickly tell us if two polygons do not overlap is useful. The minimax test will do
just that. This test says that if we place boxes around two polygons and if the two
boxes don’t overlap, then the polygons within them cannot overlap. (See Figure 9-9.)
We want the boxes to be as small as possible and still contain the polygons. That
means we want the top of the box to be as low as possible. The lowest we can make the
top of the box is where it just touches the highest point on the polygon, that is, the

Minimum y

Minimum x Maximum x

FIGURE 9-9
The minimax test. If the boxes do not overlap,
then the polygons cannot overlap.
 HIDDEN SURFACES AND LINES 323

maximum of the y coordinates of the polygon’s vertices. Similarly, the highest position
of the bottom of the box is the minimum of the y coordinates of the vertices. Two
boxes will not overlap if one is above the other, that is, if the bottom of one box is
higher than the top of the other box. This means that the minimum of the y coordinates
of one polygon is greater than the maximum of the y coordinates of the other polygon.
We can, of course, switch the roles of the two polygons. We can also compare left and
right sides by a similar test using the x coordinates.
If the minimax test is true, then the polygons do not overlap; but if the boxes do
overlap, we still are not certain whether the polygons within them overlap. We will
have to perform further tests.
A minimax test applied to the z coordinates can often tell the relative ordering of
two polygons which do overlap in x and y. If the smallest z value for one polygon is
larger than the largest z value for the other polygon, then the first polygon lies in front.
Another test which is useful is to see if all the vertices of one polygon lie on the
same side of the plane containing the other polygon. If polygon P has all its vertices on
the same side of the plane of polygon Q as the viewer, then Pis in front of Q. If all the
vertices lie in the other half-space, then P is behind Q. (See Figure 9-10.)
This test may also be inconclusive because the plane of polygon Q may intersect
polygon P. If this should happen, we can try comparing the vertices of Q against the
plane of P. Again, this may or may not yield results. (See Figure 9-11.)
If the above tests fail, we might break up the polygons into pieces such that the
tests succeed, or we might try to find an xy point common to both polygons and com-
pare the corresponding z-coordinate values.

WARNOCK’S ALGORITHM
An interesting approach to the hidden-surface problem was presented by Warnock
[WAR69]. His method does not try to decide exactly what is happening in the scene
but rather just tries to get the display right. As the resolution of the display increases,
the amount of work which the algorithm must do to get the scene right also increases.

FIGURE 9-10
Vertices of one polygon on one side of the plane of
another polygon.
324 CHAPTER NINE

FIGURE 9-11
Vertices of each polygon straddle the plane of the other.

(This is also true for scan-line algorithms.) The algorithm divides the screen up into
sample areas. In some sample areas it will be easy to decide what to do. If there are no
faces within the area, then it is left blank. If the nearest polygon completely covers it,
then it can be filled in with the color of that polygon. If neither of these conditions
holds, then the algorithm subdivides the sample area into smaller sample areas and
considers each of them in turn. This process is repeated as needed. (See Figure 9-12.)
It stops when the sample area satisfies one of the two simple cases or when the sample
area is only a single pixel (which can be given the color of the foremost polygon). The
process can also be allowed to continue to half or quarter pixel-sized sample areas,
whose color may be averaged over a pixel to provide antialiasing.
The test for whether a polygon surrounds or is disjoint from the sample area is
much like a clipping test to see if the polygon sides cross the sample-area boundaries.

FIGURE 9-12
Subdivision of a scene.
 HIDDEN SURFACES AND LINES 325

Actually, the minimax test can be employed to identify many of the disjoint polygons.
A simple test for whether a polygon is in front of another is a comparison of the z coor-
dinates of the polygon planes at the corners of the sample area.
At each subdivision, information learned in the previous test can be used to
simplify the problem. Polygons which are disjoint from the tested sample area will also
be disjoint from all of the subareas and do not need further testing. Likewise, a poly-
gon which surrounds the sample area will also surround the subareas.

FRANKLIN ALGORITHM
We mentioned how the number of possible comparisons of polygons grows as the
square of the number of polygons in the scene. Many of the hidden-surface algorithms
exhibit this behavior and have serious performance problems on complex scenes.
Franklin [FRA80] developed an approach which gives linear time behavior for most
scenes. This is done by overlaying a grid of cells on the scene (similar to Warnocks ap-
proach, only these cells are not subdivided). The size of the cells is on the order of the
size of an edge in the scene. At each cell the algorithm looks for a covering face and
determines which edges are in front of this face. It then computes the intersections of
these edges and determines their visibility. (See Figure 9-13.) The idea is that as ob-
jects are added to the scene and the number of polygons increases, the new objects will
either be hidden by objects already in the scene or will hide other objects in the scene.
While the number of objects increases, the complexity of the final scene (after hidden
portions are removed) does not increase. By considering only the edges in front of the
covering face for a cell, the algorithm considers only the edges likely to be in the final
image. Although the total number of edges may increase, this increase occurs, for the
most part, behind the covering faces, and the number of edges in front will remain
small.

S/
Lif
LSPA
(ZL
LAL
Sf

Must consider edges in front of the covering face

FIGURE 9-13
Faces behind the first face to cover a cell can be ignored for that cell.
326 CHAPTER NINE

HIDDEN-LINE METHODS
The hidden-surface techniques such as the painter’s algorithm, which rely on the prop-
erties of the frame buffer, are not sufficient for calligraphic displays. On such displays
we must not draw the hidden portions of lines. This means that for each line, we must
decide not only what objects lie in front of it but also just how those objects hide it.
Calligraphic displays were available long before raster displays became economical,
and the hidden-line problem was attacked before the hidden-surface problem.
The first solution [ROB63] compared lines to objects. For each object, it consid-
ered relevant edges to see if the object hid them. The object might not hide an edge at
all or might hide it entirely. It might hide an end, making the visible portion of the
edge smaller, or hide the middle, making two smaller visible line segments. After com-
parison of the line and object, the resulting visible line segments were compared in
turn to the remaining objects. A segment which survives comparison to all objects is
drawn.
We do not have to compare the line against all of the polygon edges in an object
in order to tell if the object hides the line. The only edges which can change whether
the line is visible or not are those on the boundary where a front face meets a back
face. These are called contour edges. We can find the contour edges by examining the
object. For a solid object, each edge has two polygons adjacent to it. If the polygons
which meet at the edge are both front faces or both back faces, then we have an interior
edge; but if one is a front face and the other a back face, then it is a contour edge. (See
Figure 9-14.)
Instead of comparing all lines to each object, we can compare the contour edges
of all objects to each line. For each intersection of a line with a contour edge, the line
either passes behind an object or emerges from it. So with each such intersection, the
number of faces hiding the line either increases or decreases by 1. Appel [APP67] has
termed the number of faces hiding a line its quantitative invisibility. His method for hid-
den-line removal is to find the quantitative invisibility for an initial point on a line, and
then to follow along the connected lines, finding intersections with the contour edges

io
>

Interior edge Contour edge

FIGURE 9-14
Interior and contour edges.
 HIDDEN SURFACES AND LINES 327

and maintaining the quantitative invisibility. The portions of the lines with quantitative
invisibility 0 are drawn.

BINARY SPACE PARTITION

Now let’s implement a hidden-surface algorithm. The method we shall use is based on
the painter’s algorithm and is credited to Fuchs [FUC80]. It is called binary space parti-
tion (BSP). The idea is to sort the polygons for display in back-to-front order. To order
the polygons we use the test which compares all the vertices of one polygon against the
plane of another polygon. We extend this test so that if the plane intersects the poly-
gon, we divide the polygon along the plane. Using this test we can pick one polygon
and compare all the other polygons to it, splitting them into two groups, those in front
and those behind. For each of these two subgroups we can again select a polygon and
use it to separate the subgroup. This process is repeated until all polygons have been
sorted. The sort works in the same fashion as Hoare’s Quicksort algorithm, repeatedly
separating the polygons into smaller and smaller groups. The result may be pictured as
a binary tree. At each node of the tree is a polygon. In one subtree, or branch, are all
the polygons in front of the plane of this polygon, and in the other branch are those
polygons which lie behind it. (See Figure 9-15.) Once we have created this tree, we
can perform an in-order traversal to obtain the polygons in back-to-front order.
Before discussing the details of the algorithm, we note that it requires the collec-
tion and sorting of all the polygons before any of them are drawn. We shall need some
storage area to collect the polygon instructions until we are ready to sort them. We
shall need to know when to start collecting polygons and also when all polygons have
been collected so that sorting can proceed. We shall also need to save more information
about a polygon than just its vertices. We shall need to save the edge and fill styles for

E
Vie :
yee
je

an A
Je

a ae
ax cee

FIGURE 9-15
An ordered tree of polygons.
328 CHAPTER NINE

the polygon. The change-of-style commands will have to be inserted to match the
order in which the polygons are actually drawn (not the order that they are entered by
the user). It will also be useful to save the equation for the plane of the polygon. This
can be used in the polygon comparison.
Let’s first consider when to save and when to sort the polygons. In our system
we shall limit hidden-line processing to single display-file segments. We shall sort out
all the objects within a segment, but we will not check to see if an object in one seg-
ment hides an object in a different segment. With this organization, a reasonable point
at which to sort and save the polygons is just before the display-file segment is closed.
When the segment is closed, we know that all relevant polygons have been entered and
no other display-file segment’s polygons will be around to confuse us. We shall there-
fore modify the CLOSE-SEGMENT algorithm to include a call to our hidden-surface
removal routine. We shall use the HIDDEN flag to tell us to save polygons, so poly-
gons will be saved any time this flag is TRUE; but with each CLOSE-SEGMENT, the
collected polygons will be sorted and moved to the display file. This groups them ac-
cording to segment.
The modified CLOSE-SEGMENT routine calls the HIDDEN-SURFACE-
CHECK routine to actually do the sorting and saving. It looks as follows:

9.5 Algorithm CLOSE-SEGMENT (Modification of algorithm 5.2)

Global NOW-OPEN the name of the currently open segment
FREE the index of the next free display-file cell
HIDDEN the hidden-surface removal flag
SEGMENT-START, SEGMENT-SIZE arrays for start and size of the seg-
ments
BEGIN
IF NOW-OPEN = 0 THEN RETURN ERROR ‘NO SEGMENT OPEN’;
DELETE-SEGMENT(0);
IF HIDDEN THEN HIDDEN-SURFACE-CHECK;
SEGMENT-START[0] <— FREE;
SEGMENT-SIZE[0] < 0;
NOW-OPEN < 0;
RETURN;
END;

To save the polygons we shall use some new arrays. We shall use arrays ID, XD,
YD, and ZD to hold the polygon edge instructions (the polygon-vertices). We shall call
this the D buffer. The arrays must be large enough to hold all of the vertices of all of
the polygons being sorted. We shall also use arrays POLY-START, POLY-SIDES,
POLY-FILL-STYLE, POLY-EDGE-STYLE, POLY-A, POLY-B, POLY-C, POLY-D,
POLY-FRONT, and POLY-BACK to form a table of information about each polygon.
There is one entry for each polygon drawn. POLY-START will indicate where in the D
buffer the vertices for the polygon may be found. POLY-SIDES is the number of edges
on the polygon. POLY-FILL-STYLE and POLY-EDGE-STYLE remember the style of
the polygon. The arrays POLY-A, POLY-B, POLY-C, and POLY-D contain the coeffi-
cients for the equation of the plane of the polygon. POLY-FRONT and POLY-BACK
 HIDDEN SURFACES AND LINES 329

will hold the indices of other polygon entries. They are the branches that will form our
sorted tree of polygons.
We shall modify our BACK-FACE-CHECK algorithm to save the polygons. The
modifications will be to calculate the entire normal vector to the polygon plane (not
just its z component) and to check the solid flag, calling the SAVE-POLY-FOR-HSC
routine if the polygon is filled.

9.6 Algorithm BACK-FACE-CHECK(POLYSIZE) (Revision of algorithm 9.3) Filters

out polygon drawn clockwise
Argument POLYSIZE the number of sides on the polygon
Global XT, YT, ZT T-buffer array storage of the vertex points
SOLID a flag which indicates polygon filling
Local A,
B, Ca vector for the normal to the plane of the polygon
I, J for stepping through the vertices
BEGIN
A<— 0;
B <0;
C<—0;
FOR I = 1 TO POLYSIZE DO
BEGIN
IFI = POLYSIZE THEN J < 1
ELSE J <—I+ 1;
A<A + ((YT{I] — YT{J)) « (ZTI + ZT);
, B<—B + ((ZT[I] — ZT[J)) * (XTIN) + XT[J)):
C<C + ((XTII] — XT) *« (YTI] + YT);
END;
IF C < 0 THEN RETURN;
IF SOLID THEN SAVE-POLY-FOR-HSC(POLY SIZE, A, B, C)
ELSE
~~
BEGIN
VIEWING-TRANSFORM(POLYSIZE, XT[POLYSIZE], YT[POLYSIZE]);
FOR I = 1 TO POLYSIZE DO VIEWING-TRANSFORM(IT[I], XT[I],
YT[I});
END;
RETURN;
END;

The SAVE-POLY-FOR-HSC algorithm enters the polygon information into the

polygon table at POLY and into the D buffer starting at DFREE. Most of the entries are
straightforward. As we saw in Chapter 8, the A, B, and C coefficients in the equation
of the plane (Equation 8.4) can be the elements of the normal vector which is deter-
mined in BACK-FACE-CHECK. The fourth coefficient D may be found by substitut-
ing for x, y, and z in the equation the coordinates of a point known to be on the plane.
This is what is done for the POLY-D array.
By setting POLY-FRONT to POLY — 1, we formalinked list of all the polygons
(each linked to its predecessor). We shall use the linked list to access the polygons
when sorting.
330 CHAPTER NINE

9.7 Algorithm SAVE-POLY-FOR-HSC(POLYSIZE, A, B, C) Save a polygon in the

polygon table for the hidden-surface check
Arguments POLYSIZE the number of sides on the polygon
A, B, C a vector normal to the plane of the potygon
Global POLY index of the next free polygon-table cell
CURRENT-FILL-STYLE the interior style of the polygon
CURRENT-LINE-STYLE the edge style of the polygon
IT, XT, YT, ZT arrays containing the vertices of one polygon
ID, XD, YD, ZD D-buffer arrays containing the vertices of all polygons
involved in the hidden-surface check
DFREE the next free D-buffer cell
POLY-START, POLY-SIDES, POLY-FILL-STYLE, POLY-EDGE-STYLE,
POLY-A, POLY-B, POLY-C, POLY-D, POLY-FRONT, POLY-BACK the
polygon table
Local I for stepping through the polygon vertices
Constant © POLYGON-TABLE-SIZE the size of the polygon-table arrays
D-BUFFER-SIZE the size of the D buffer
BEGIN
IF POLY = POLYGON-TABLE-SIZE THEN
RETURN ERROR ‘POLYGON TABLE OVERFLOW’;
POLY-START[POLY] < DFREE;
POLY-SIDES[POLY] < POLYSIZE;
POLY-FILL-STYLE[POLY] <- CURRENT-FILL-STYLE;
POLY-EDGE-STYLE[POLY] <- CURRENT-LINE-STYLE;
POLY-A[POLY] < A;
POLY-B[POLY] < B;
POERY-C[EBOLY)]<—'¢:
POLY-D[POLY] <— — (A * XT[1] + B * YT[1] + C * ZT[I]);
POLY-FRONT < POLY — 1;
POLY-BACK < 0;
POLY <— POLY + 1;
FOR I = 1 TO POLYSIZE DO
BEGIN
IF DFREE > D-BUFFER-SIZE THEN
RETURN ERROR ‘D BUFFER OVERFLOW’;
ID[DFREE] : = IT[I);
XD[DFREE] := XT[I];
YD[DFREE] := YT{I];
ZD[DFREE] := ZT[I];
DFREE := DFREE + 1;
END;
RETURN;
END;

Now let’s consider what happens when a display-file segment is closed and
the
HIDDEN-SURFACE-CHECK algorithm is called. This procedure first determine
s if
there is anything to sort and save. If so, it picks one of the polygons to start the
sorting
process. This polygon will be the root of the sorted tree of polygons. The
algorithm
 HIDDEN SURFACES AND LINES 331

sorts the polygons, saves them in the display file in back-to-front order, and then resets
the polygon table and D buffer so that future polygons will be sorted as a new set.

9.8 Algorithm HIDDEN-SURFACE-CHECK Aroutine to remove hidden surfaces

Global POLY index of next free cell in the polygon table
SOLID the polygon-filling flag
DFREE the next free D-buffer cell
Local ROOTPOLY the root of the space partition tree
BEGIN
IF POLY = 1 THEN RETURN;
ROOTPOLY < POLY -—1;
SORT-POLYGONS(ROOTPOLY);
SAVE-POLYGONS-IN-ORDER(ROOTPOLY);
}20)Bec ile
DEREE <r
RETURN;
END;

The sorting routine is given in recursive form (it calls itself). Nonrecursive ver-
sions are possible, but they are more complex. The algorithm checks to see if there is
something to sort; if not, it returns. It uses a loop to step through the polygons in a
linked list. The index of the next polygon in the list is found in the POLY-FRONT array
for the current polygon. It compares the first polygon in the list (ROOT-NODE)
against all of the other polygons in the list (TEST-POLY). The COMPARE-POLYS pro-
cedure not only decides on which side of ROOT-NODE each TEST-POLY lies but also
links the TEST-POLY onto a sublist accordingly. When the loop has finished, the list
will have been divided into two sublists. There will be a list of the polygons in front of
the ROOT-NODE, which will be attached to the ROOT-NODE’s POLY-FRONT entry,
and there will be a list of those behind, attached to POLY-BACK for the ROOT-NODE.
The algorithm, then, recursively sorts each of these two sublists.

9.9 Algorithm SORT-POLYGONS(ROOT-NODE) A routine to build a sorted binary

"tree of polygons
Argument ROOT-NODE the polygon at the root of the tree
Global POLY-FRONT, POLY-BACK links for the tree branches
Local TEST-POLY, NEXT-POLY polygons to be compared to the root
BEGIN
IF ROOT-NODE = 0 THEN RETURN;
TEST-POLY <— POLY-FRONT[ROOT-NODE];
POLY-FRONT[ROOT-NODE] < 0;
WHILE TEST-POLY # 0 DO
BEGIN
NEXT-POLY < POLY-FRONT[TEST-POLY];
COMPARE-POLY S(ROOT-NODE, TEST-POLY);
TEST-POLY < NEXT-POLY;
END;
332 CHAPTER NINE

SORT-POLYGONS(POLY-FRONT[ROOT-NODE]});
SORT-POLYGONS(POLY-BACK[ROOT-NODE)});
RETURN;
END;

There is actually a lot of work that goes on in the COMPARE-POLYS algorithm.

It compares each vertex of the TEST polygon against the plane of the ROOT polygon
and decides which side it is on. But it will also split the TEST polygon into two sepa-
rate polygons if it is intersected by the plane. The technique is very similar to what we
used for clipping. When clipping, we compare points against a clipping plane and only
keep those points which lie on one side. In COMPARE-POLYS we keep the points on
both sides but in two separate groups, so we end up with two polygons. As with clip-
ping, when an edge crosses the plane, we calculate the intersection point, using it to
build an “‘invisible’’ edge along the plane boundary. (See Figure 9-16.)
The algorithm determines on which side of the plane a point lies by checking the
sign of the result of substituting the coordinates of the point into the expression for
the plane (see Equations 8.69 and 8.70). It finds the first vertex point which is not on
the plane and then compares following vertices, checking for a change in sign which
indicates a crossing of the plane. If a side ever crosses the plane, then we will have to
split the polygon. To do this, we form a new polygon starting at DFREE (the next free
cell in the D buffer). The new polygon begins with the first intersection point. We must
also save the points for the other side of the split. This side will also include the inter-
section point. Since the split polygon can have more vertices that the original, we can-
not store them in place, so we use arrays IE, XE, YE, and ZE (the E buffer) to save the
other half of the split polygon. The E buffer must be large enough to hold one poly-
gon. After the first intersection is discovered, vertices are copied to either the D-buffer
piece or the E-buffer piece. Intersection points go to both.

FIGURE 9-16
Dividing the polygon alonga plane.
 HIDDEN SURFACES AND LINES 333

Although we are finding the sign of each vertex, we are actually looking for
changes in sign between two vertices; that is, we are considering the edge between the
vertex pair. After stepping through all vertices, we must consider the first once again in
order to characterize the final edge, that is, to close the polygon.

9.10 Algorithm COMPARE-POLYS(ROOT, TEST) Finds on which side of ROOT the

TEST polygon lies and links it to that branch. If the plane of ROOT intersects the TEST
polygon, then it is split
Arguments ROOT the polygon providing the plane used in the comparison
TEST the polygon being compared and linked (and possibly split)
Global POLY-START, POLY-SIDES, POLY-A, POLY-B, POLY-C, POLY-D,
POLY-FRONT the polygon table
ID, XD, YD, ZD the D-buffer arrays containing polygon vertices
DFREE the next free D-buffer cell
IE, XE, YE, ZE the E-buffer arrays for vertices of split-off polygon
EFREE the next free E-buffer cell
Local FIRST-V index of the first vertex of the polygon
LAST-V index of the last vertex of the polygon
CROSS-V index of the vertex just before the polygon crosses the test plane
SPLIT-V index of the first vertex of the split polygon
J for stepping through the polygon vertices
STEST the first nonzero sign parameter
SLAST the sign parameter of the previous vertex
SNEW the sign parameter for the current vertex
UO, U, V test point parameters
W parameter for the intersection point
XP, YP, ZP the intersection point
CHANGED-BEFORE aflag to indicate the first time the polygon crosses
the plane
CROSSED aflag to indicate if the polygon edge crossed the plane
BEGIN
FIRST-V <— POLY-START[TEST];
LAST-V < FIRST-V + POLY-SIDES[TEST] — 1;
CROSS-V < FIRST-V;
UO <— POLY-A[ROOT] * XD[FIRST-V] + POLY-B[ROOT] * YD[FIRST-V]
+ POLY-C[ROOT] « ZD[FIRST-V] + POLY-D[ROOT];
STEST <— SIGNOF[U0];
find the first vertex not on the plane
U <— U0;
WHILE STEST = 0 AND CROSS-V < LAST-V DO
BEGIN
CROSS-V < CROSS-V + 1;
U < POLY-A[ROOT] * XD[CROSS-V] + POLY-B[ROOT] * YD[CROSS-V]
+ POLY-C[ROOT] * ZD[CROSS-V] + POLY-D[ROOT],
STEST < SIGNOF(U);
END;
IF STEST = 0 THEN
'
BEGIN
test polygon was contained in the plane so treat as in front
334 CHAPTER NINE

POLY-FRONT[TEST] <- POLY-FRONT[ROOT];

POLY-FRONT[ROOT] < TEST;
RETURN;

SLAST < STEST;

EFREE < 1;
SPLIT-V — DFREE;
CHANGED-BEFORE < FALSE;
FOR J = CROSS-V + 1 TO LAST-V DO
BEGIN
step through all vertices
V — POLY-A[ROOT] * XD[J] + POLY-B[ROOT] * YD{J]
+ POLY-C[ROOT] * ZD[J] + POLY-D[ROOT];
SNEW < SIGNOF(V);
IF SNEW = — SLAST THEN
BEGIN
point has crossed the plane, find intersection
CROSSED < TRUE;
W<U/(U —- V);
XP <—(XD{J] — XD[J — 1])*W+ XD{J — 1);
YP — (YD(J] — YD[J — 1])*W + YD[J — 1);
ZP <—(ZD[J] — ZD[J — 1])*W + ZD{J — 1);
CALL PUT-IN-D(UID[J], XP, YP, ZP);
CALL PUT-IN-E(ID[J], XP, YP, ZP);
SLAST <— SNEW;
END y
ELSE CROSSED < FALSE;
IF SNEW = STEST OR (SNEW = 0 AND SLAST = STEST) THEN
BEGIN
point belongs on the first side of the plane
IF CROSSED THEN IE[EFREE — 1] <1;
IF CHANGED-BEFORE THEN PUT-IN-E(ID{[J], XD{J], YD[J], ZD[J})
ELSE CROSS-V <— CROSS-V + 1;
END;
ELSE
BEGIN
point belongs on the second side of the plane
CHANGED-BEFORE < TRUE;
IF CROSSED THEN ID[DFREE — 1] — 1;
CALL PUT-IN-D(ID[J], XD[J], YD[J}, ZD[J]);
END;
U<YV;
END;
now we must close the polygon
IF SLAST ¥ STEST THEN
BEGIN
W<V/(V — UO);
XP <— (XD[FIRST-V] — XD[{J]) « W + XD{[J];
YP <— (YD[FIRST-V] — YD[J]) * W + YD[J];
ZP <— (ZD[FIRST-V] — ZD[J]) *W + ZD{J);
 HIDDEN SURFACES AND LINES 335

CALL PUT-IN-D(ID[FIRST-V], XP, YP, ZP);

CALL PUT-IN-E(1, XP, YP, ZP);
END;
finally, hang the polygon(s) on the ROOT node
IF SPLIT-V DFREE THEN SET-UP-SPLIT-POLY (ROOT, TEST, SPLIT-V, STEST);
ADJUST-ORIGINAL-POLY (ROOT, TEST, FIRST-V, CROSS-V, STEST);
RETURN;
END;

Once the polygon has been tested (and split if necessary), two routines are used
to clean things up. If the polygon is split, then the vertices for one half are copied in
the D buffer. But we do not have a corresponding polygon-table entry. The routine
SET-UP-SPLIT-POLY forms this table entry. It also links the polygon onto either the
POLY-FRONT or POLY-BACK branch of the ROOT polygon, depending on whether it
lies in front or in back of the ROOT polygon plane.

9.11 Algorithm SET-UP-SPLIT-POLY(ROOT, TEST, SPLIT-V, STEST) Completes

a polygon-table entry for the split-off polygon and attaches it to the root polygon
Arguments ROOT the root polygon
TEST the original test polygon
SPLIT-V index of the first vertex of the split-off polygon
STEST indicates on which side of the root the split polygon lies
Global POLY-START, POLY-SIDES, POLY-FILL-STYLE, POLY-EDGE-STYLE,
POLY-A, POLY-B, POLY-C, POLY-D, POLY-FRONT, POLY-BACK the
polygon table
POLY index of the next free polygon-table cell
BEGIN
POLY-START[POLY] <— SPLIT-V;
POLY-SIDES[POLY] < DFREE — SPLIT-V;
POLY-FILL-STYLE[POLY] <— POLY-FILL-STYLE[TEST];
POLY-EDGE-STYLE[POLY] <— POLY-EDGE-STYLE[TEST];
POLY-A[POLY] <— POLY-A[TEST];
POLY-B[POLY] <— POLY-B[TEST];
POLY-C[POLY] < POLY-C[TEST];
, POLY-D[POLY] <— POLY-D[TEST];
POLY-BACK[POLY] < 0;
IF STEST < 0 THEN
BEGIN
POLY-FRONT[POLY] <- POLY-FRONT[ROOT];
POLY-FRONT[ROOT] < POLY;
END
ELSE
BEGIN
POLY-FRONT[POLY] <— POLY-BACK[ROOT];
POLY-BACK[ROOT] < POLY;
END;
POLY < POLY + 1;
RETURN;
END;
336 CHAPTER NINE

ee)
Original | CROSSV
polygon

Split C
piece DFREE
D5.
N DFREE

E Buffer

Before After
FIGURE 9-17
Case where the number of sides increases with the split.

The second cleanup routine is called ADJUST-ORIGINAL-POLY. If the polygon

was split, then the points for one of the subpolygons may be found partly in the D buf-
fer and partly in the E buffer. The D buffer holds the vertices up to the first intersection
with the ROOT polygon plane. The remainder is in the E buffer. These two parts must
be recombined. If this subpolygon now has fewer vertices than the original, then the
vertices in the E buffer can be copied back into the original D-buffer locations. If,
however, the size of the polygon has grown, then both the D-buffer piece and the E-
buffer piece are copied to a new D-buffer location. The result is linked onto the appro-
priate branch of the ROOT polygon. (See Figures 9-17 and 9-18.)

9.12 Algorithm ADJUST-ORIGINAL-POLY

(ROOT, TEST, FIRST-V, CROSS-V,
STEST) Combines all vertices of the polygon in the D buffer and attaches it to the root
polygon
Arguments ROOT the root polygon
TEST the original test polygon
FIRST-V index of the first vertex of the polygon
CROSS-V the vertex just before the polygon crosses the test plane
STEST indicates on which side of the root the split polygon lies
 HIDDEN SURFACES AND LINES 337

D Buffer

2s:
FIRSTV.

<
Split [—
a 2
piece L. DFREE DFREE

E Buffer
EFREE \

Before After
FIGURE 9-18
Case where the number of sides does not increase with the split.

Global POLY-START, POLY-SIDES, POLY-FRONT, POLY-BACK from the polygon

table
ID, XD, YD, ZD the D-buffer arrays
DFREE the next free D-buffer cell
IE, XE, YE, ZE the E-buffer arrays
EFREE the next free E-buffer cell
Local J an index for stepping through the vertices
SIDES the number of sides on the polygon
BEGIN
SIDES < EFREE + CROSS-V — FIRST-V;
IF SIDES <= POLY-SIDES[TEST] THEN
BEGIN
the adjusted polygon will fit in the original D-buffer space
FORJ = 1 TO EFREE — 1 DO
BEGIN
ID[CROSS-V + J] <IE[J];
XD[CROSS-V + J] — XE[{J];
YD[CROSS-V + J] — YE[J];
338 CHAPTER NINE

ZD[CROSS-V + J] <— ZE[J];

END;
END
ELSE .
BEGIN
POLY-START[TEST] < DFREE;
FOR J = FIRST-V TO CROSS-V DO PUT-IN-D(ID[J], XD[J], YD[J], ZD{J));
FOR J = 1TO EFREE DO PUT-IN-DUE[J], XE[J], YE[J], ZE[J]);
END;
POLY-SIDES[TEST] < SIDES;
IF STEST < 0 THEN
BEGIN
POLY-FRONT[TEST] <— POLY-BACK[ROOT];
POLY-BACK[ROOT] < TEST;
END
ELSE
BEGIN
POLY-FRONT[TEST] < POLY-FRONT[ROOT];
POLY-FRONT[ROOT] < TEST;
END;
RETURN;
END;

The comparison algorithms use the following utility routines to characterize the
sign of a number and to save a-vertex in the D buffer or E buffer.

9.13 Algorithm SIGNOF(X) Returns — 1, 0, or 1 to indicate the sign of X

Argument Xa value from which the sign is to be extracted
Constant | ROUNDOFF some small number greater than any round-off error
BEGIN
SIGNOF < 0;
IF X < —ROUNDOFF THEN SIGNOF < —1;
IF X > ROUNDOFF THEN SIGNOF < 1;
RETURN;
END;

9.14 Algorithm PUT-IN-D(OP, X, Y, Z) Saves an instruction in:the D buffer

Arguments OP, X, Y, Z a display-file instruction
Global ID, XD, YD, ZD the D-buffer arrays
DFREE the next free D-buffer cell
BEGIN
ID[DFREE] <— OP;
XD[DFREE] < X;
YD[DFREE] < Y;
ZD[DFREE] < Z;
DFREE < DFREE + 1;
RETURN;
END;
 HIDDEN SURFACES AND LINES 339

9.15 Algorithm PUT-IN-E(OP, X, Y, Z) Saves an instruction in the E buffer

Arguments OP, X, Y, Z a display-file instruction
Global IE, XE, YE, ZE the E-buffer arrays
EFREE the next free E-buffer cell
BEGIN
IE[EFREE] <— OP;
XE[EFREE] < X;
YE[EFREE] < Y;
ZE[EFREE] < Z;
EFREE < EFREE + 1;
RETURN;
END;

Once we have the sorted tree of polygons, we can traverse it to enumerate them
in back-to-front order. This is an in-order traversal of the tree which is most simply ex-
pressed as the following recursive algorithm.

9.16 Algorithm SAVE-POLYGONS-IN-ORDER(ROOT) Enter polygons into the

display file in back-to-front order
Argument ROOT the root polygon of the spatially sorted tree
Global POLY-FRONT, POLY-BACK arrays of the polygon table for the branch links
BEGIN
IF ROOT = 0 THEN RETURN;
SAVE-POLYGONS-IN-ORDER(POLY-BACK[ROOT]);
SEND-TO-DF(ROOT);
SAVE-POLYGONS-IN-ORDER(POLY-FRONT[ROOT]);
RETURN;
END;

The algorithm first checks that there is something in the tree. If so, it recursively
enters into the display file (in back-to-front order) all of the polygons behind the
ROOT polygon. It next enters the ROOT polygon into the display file. Finally, it enters
all of the polygons in front of the ROOT polygon by means of another recursive call.
In SEND-TO-DF we first make sure that the proper edge and fill styles are in ef-
fect. Then we enter the polygon-drawing command. Finally, we enter the polygon
sides.

9.17 Algorithm SEND-TO-DF(POLYGON) Enters a polygon into the display file

Argument POLYGON the polygon to be entered
Global POLY-START, POLY-SIDES arrays that are part of the polygon table
ID, XD, YD arrays that make up the D buffer
Local I for stepping through the polygon vertices
F the final vertex of the polygon
BEGIN
CHECK-STYLE(POLYGON),
F — POLY-START[POLYGON] + POLY-SIDES[POLYGON] — 1;
enter the polygon instruction
VIEWING-TRANSFORM(POLY-SIDES[POLYGON], XD[F], YD[F]);
340 CHAPTER NINE

enter the polygon sides

FOR I = POLY-START[POLYGON] TO F DO
VIEWING-TRANSFORM(ID[I], XD[IJ, YD[I]);
RETURN; .
END;

Checking for the proper edge and fill style is done through the CHECK-STYLE
utility.

9.18 Algorithm CHECK-STYLE(POLYGON) Makes sure that the polygon is drawn

in the correct style
Argument POLYGON the polygon to be drawn
Global CURRENT-FILL-STYLE the interior style of the polygon
CURRENT-LINE-STYLE the edge style of the polygon
POLY-FILL-STYLE, POLY-EDGE-STYLE the style entries in the polygon
table
BEGIN
IF CURRENT-LINE-STYLE # POLY-EDGE-STYLE[POLYGON] THEN
BEGIN
DISPLAY-FILE-ENTER(POLY-EDGE-STYLE[POLYGON));
CURRENT-LINE-STYLE <— POLY-EDGE-STYLE[POLYGON];
END;
IF CURRENT-FILL-STYLE # POLY-FILL-STYLE[POLYGON] THEN
BEGIN
DISPLAY-FILE-ENTER(POLY-FILL-ST YLE[POLYGON]);
CURRENT-FILL-STYLE <— POLY-FILL-STYLE[POLYGON];
END;
RETURN;
END;

All that remains for our hidden-surface check is an initialization procedure. We

use it to set the default edge and fill styles, and we empty the D buffer and polygon
table.

9.19 Algorithm INITIALIZE-9 Initialization for hidden-surface routines

Global CURRENT-FILL-STYLE the interior style of the polygon
CURRENT-LINE-STYLE the edge style of the polygon
DFREE the next free D-buffer cell Ys
POLY index of the next free polygon-table cell
BEGIN
INITIALIZE-9A;
CURRENT-LINE-STYLE < 0;
CURRENT-FILL-STYLE <— — 16;
POLY <— 1;
DFREE < 1;
RETURN;
END;

This completes the algorithms needed for hidden-surface removal. Our graphics
system does not take full advantage of the power of the binary space partition. With
 HIDDEN SURFACES AND LINES 34]

each change of viewing parameters we require the applications program to redraw the
scene, constructing fresh display-file segments and causing us to resort the polygons.
But note that if only the view changes, and not the objects in the scene, then the sort is
essentially the same each time. Different views may change whether a sublist of poly-
gons is in front or in back of the root polygon, but it will not change the way that the
polygons were divided into sublists. For some applications which show different views
of a fixed world, we can take advantage of this, doing the sort only once, building a
single tree. This tree is then traversed in the appropriate order for each scene. To do
this, the sorted tree and hidden-surface removal must be incorporated into the applica-
tions program.

AN APPLICATION
A major application of computer graphics is in computer-aided design (CAD). The
computer can aid in the design process in many areas from architecture to machine
parts to electronic circuits. The machine can maintain a data base describing the object
being designed. Design components and properties of mating parts may also be avail-
able. A model for the part can be constructed and its behavior studied via computer
simulation. Of course, it is often quite helpful to be able to draw and to view the part,
and this is where computer graphics comes in.
Consider a computer-aided design system for designing machine parts. Such a
system might provide a set of primitive shapes and operations for combining them to
produee the desired object. But a complex object, if shown in wire frame, may be too
confusing. To make a realistic-looking object we should remove hidden lines and sur-
faces. We should also be able to view the object from different directions so as to in-
spect all sides of it. We might even employ clipping planes to inspect the interior. To do
this on our system we would first set the viewing and clipping parameter for the de-
sired view. We would next turn on the HIDDEN flag and open a display-file segment.
We would then draw the part using POLYGON-ABS-3 and POLYGON-REL-3 calls.
Finally, we would close and display the segment. The hidden-surface software will pro-
duce a realistic image.

FURTHER READING
The first solution to the hidden-line problem was described in [ROB63]. Other hidden-
line solutions are described in [APP67], [GAL69], and [LOU70]. Back-face removal
is discussed in [LOU70]. The Z buffer is described in [CAT74]. A scan-line Z buffer is
suggested in [CAR76]. Other scan-line algorithms are described in [BOU70],
[GEA77], and [WYL67]. Some of the hidden-surface algorithms first decompose the
polygons into triangles or trapezoids. This ensures that the objects are convex and pro-
vides a uniformity which makes them easier to manipulate. Methods for this decom-
position are described in [JAC80] and [LIT79]. Sorting into priority order and the
painter’s algorithm are described in [NEW72]. Warnock’s algorithm may be found in
[WAR69]. Binary space partition is presented in [FUC80]. Its use for fast generation
of different views of a scene is described in [FUC83]. An excellent description and
comparison of hidden-surface techniques are found in [SUT74], with a shorter version
342 CHAPTER NINE

in [SUT73]. A review of some hidden-surface methods and a discussion of how hierar-

chical structuring of objects can improve quality and performance -are given in
[CLA76]. A bibliography of hidden-line and hidden- surface papers of the 1960s and
1970s appears in [GRI78].

[APP67] Appel, A. ‘‘The Notion of Quantitative Invisibility and the Machine Rendering of Solids,’’ Pro-
ceedings of the ACM National Conference, pp. 387-393 (1967).
[BOU70] Bouknight, W. J., ‘“‘A Procedure for Generation of Three Dimensional Half-Tone Computer
Graphics Representations,’’ Communications of the ACM, vol. 13, no. 9, pp. 527-536 (1970).
[CAR76] Carpenter, L., ‘‘A New Hidden Surface Algorithm,’’ Proceedings of NW76, ACM, Seattle, WA,
(1976).
[CAT74] Catmull, E., ‘‘A Subdivision Algorithm for Computer Display of Curved Surfaces,’’ University of
Utah, Salt Lake City (Dec. 1974).
[CLA76] Clark, J. H., ‘Hierarchical Geometric Models for Visible Surface Algorithms,’’ Communications
of the ACM, vol. 19, no. 10, pp. 547-554 (1976).
[FUC80] Fuchs, H., Kadem, Z., ‘‘On Visible Surface Generation by a Priori Tree Structures,’’ Computer
Graphics, vol. 14, no. 3, pp. 124-133 (1980).
[FUC83] Fuchs, H., ‘‘Near Real-Time Shaded Display of Rigid Objects,’’ Computer Graphics, vol. 17, no.
3, pp. 65-72 (1983).
[GAL69] Galimberti, R., Montanari, U., “‘An Algorithm for Hidden-Line Elimination ,’’ Communications of
the ACM, vol. 12, no. 4, pp. 206-211 (1969).
[GRI78] Griffiths, J. G., “‘Bibliography of Hidden-Line and Hidden-Surface Algorithms,’’ Computer Aided
Design, vol. 10, no. 3, pp. 203-206 (1978).
{[HAM77] Hamlin, G., Gear, C. W., ‘‘Raster-Scan Hidden Surface Algorithm Techniques,’’ Computer
Graphics, vol. 11, no. 2, pp. 206-213 (1977).
[JAC80] Jackson, J. H., ‘‘Dynamic Scan-Converted Images with a Frame Buffer Display Device,’’ Comput-
er Graphics, vol. 14, no. 3, pp. 163-169 (1980).
[LIT79] Little, W. D., and Heuft, R., “‘An Area Shading Graphics Display System,’’ JEEE Transactions on
Computers, vol. C-28, no. 7, pp. 528-531 (1979).
{LOU70] Loutrel, P. P., “‘A Solution to the Hidden-Line Problem for Computer-Drawn Polyhedra,’’ EEE
Transactions on Computers, vol. C-19, no. 3, pp. 205-213 (1970)
[NEW72] Newell, M. E., Newell, R. G., Sancha, T. L., “‘A New Approach to the Shaded Picture Prob-
lem,”’ Proceedings of the ACM National Conference (1972).
[ROB63] Roberts, L. G., ‘‘Machine Perception for Three-Dimensional Solids,’’ MIT Lincoln Laboratory,
TR 315 (May 1963).
[SUT73] Sutherland, I. E., Sproull, R. EK, Schumacker, R. A., ‘‘Sorting and the Hidden-Surface Problem,’’
Proceedings of the National Computer Conference, vol. 42, pp. 685-693, AFIPS Press (1973).
[SUT74] Sutherland, I. E., Sproull, R. EF, Schumacker, R. A., ‘“‘A Characterization of Ten Hidden-Surface
Algorithms,’’ Computing Surveys, vol. 6, no. 1, pp. 1-55 (1974).
[WAR69] Warnock, J. E., ‘‘A Hidden-Surface Algorithm for Computer-Generated Halftone Pictures,’’ Com-
puter Science Department, University of Utah, TR 4-15 (1969). Se
[WYL67] Wylie, C., Romney, G., Evans, D., Erdahl, A., ‘‘Half-Tone Perspective Drawings by Computer,”’
AFIPS FJCC, vol. 31, pp. 49-58 (1967).

EXERCISES
9-1 For each of the following triangles give a vector normal to the plane which contains it.
a) O12) 5 Qalie (Ber aen 2)
b)43 331) i253, Gass)
c) (, 1, 1), C1, 0, 1), ©, 0, 0)
a) 6527 1) (Sa) eee)
e) (—5, 3, 7), 4, —2; 1); 0, 6; —2)
 HIDDEN SURFACES AND LINES 343

9-2 Consider the vectors

A=f[l 0 0]
B=[0 1 0)
Cail eh ol
D=(2°3 4]
E=[(-4 2 -3]
Calculate the following dot and cross products:
A»B,A=D;>B-D,C-D,D:E
AXB,AxXC,AxD,BxC,BxD,CxD,DxC,DXxE,ExD
9-3 Suppose that POLYGON-ABS-3 commands are issued for the triangles in Exercises 9-1,
with the argument arrays containing the points in the same order as listed above. Under the
default viewing parameters, state for each triangle whether it is a front or a back face.
9-4 Show that the sums in Equation 9.3 do compute twice the area of the projections of the
polygon in the x, y, and z directions. Hint: Consider the area of the triangle between vertex i,
vertex j, and the origin.
9-5 Some special cases which cause problems for some hidden-surface algorithms are penetrat-
ing faces and cyclic overlap. A penetrating face occurs when polygon A passes through polygon
B. Cyclic overlap occurs when polygon Ais in front of polygon B, which is in front of polygon
C, which is in front of Polygon A. Actually, we need only two polygons for cyclic overlap; im-
agine a rectangle threaded through a polygon shaped like the letter C so that it is behind the top
of the C but in front of the bottom part. For the various hidden-surface methods we have pre-
sented, discuss whether or not they can handle penetrating faces and cyclic overlap.
9-6 a) Show that no polygon subdivision takes place in applying the binary space partition
method to a convex object.
b) For the case of a convex object compare the cost of the back-face removal method
with that of the binary space partition method for a single view.
c) Suppose we wish to display a sequence of views of a convex object. How would the
cost of using back-face removal compare to the binary space partition scheme?

PROGRAMMING PROBLEMS
9-1 Implement algorithms 9.1 through 9.4 to extend the graphics system for back-face removal.
9-2 Test the back-face removal algorithm by drawing a house built from unfilled polygons.
Select a single view which shows front, side, and roof.
9-3 Test the back-face removal algorithm by drawing a die. Provide different views of the die so
that each face is displayed at least once.
9-4 Modify the back-face algorithm for unfilled polygons so that instead of removing back faces
it draws them in a less pronounced line style (e.g., as dashed lines).
9-5 Implement algorithms 9.5 through 9.19 to provide the painter’s algorithm for hidden-surface
removal.
9-6 Test the painter’s algorithm by showing several filled polygons with different interior styles,
and different states of overlap, entered in mixed order.
9-7 Test the painter’s algorithm by showing two houses composed of filled polygons with differ-
ent interior styles. Select a view such that one house partially obscures the other house.
344 CHAPTER NINE

9-8 Write a program which gives successive views of an object drawn by procedure FOO within
the cube 0 < x < 1,0 <y < 1,0 < z < 1. The images should be those seen by the viewer as he
walks around the object.
*9-9 Functions of the form y = F(x, z) may be shown by a number of polygons composed of
the curve at fixed z for sample x and the boundaries. For example, a polygon might connect
points for z = 0.5 andx = 0.0, 0.2, 0.4, 0.6, 0.8, 1.0, and the boundaries at x = 1, y = O, and
x = 0. In this manner, each polygon shows aslice of the function. Several slices may be shown
by displaying polygons for several z values.
a) Write a program which does this for a curve such as

y = sin(m * x) + cos(t * Z)

b) Use the hidden-surface removal algorithm to remove portions hidden by nearer slices.
(You can get the effect of hidden-line removal by using polygons filled with white having black
edges.)

c) Select viewing parameters which best display the surface.

**9-10 Modify your graphics system to use a Z buffer to remove hidden surfaces.
 CHAPTER

TEN
LIGHT,
COLOR,

SHADING

INTRODUCTION
In previous chapters, we have seen how to construct three-dimensional objects out of
polygons. We can now project and display perspective images of objects to get a feel-
ing of depth. We have also learned how to remove hidden lines and surfaces to give the
images a greater degree of realism. The polygons from which the objects are created
can be filled in. Often, we will be able to draw polygons with different interior styles,
and different styles may give the effect of different intensities or shades for the poly-
gon surfaces. In this chapter we shall consider the shading of three-dimensional ob-
jects. We shall learn how to automatically set the polygon interior styles to give further
realism to the image. We shall develop a model for the manner in which light sources
illuminate objects, and use this model to determine how bright each polygon face
should be. We shall also discuss how colors are described and how the model may be
extended to colored objects.

DIFFUSE ILLUMINATION
Let us begin our discussion of illumination by considering an indirect lighting source.
We will assume that our object is illuminated by light which does not come from any

345
346 cCHAPTERTEN

particular source but which comes from all directions. This is background light, light
which is reflected from walls, floor, and ceiling. We assume that this light is uniform,
that there is the same amount everywhere. There will be no bright spots, and no par-
ticular direction will be favored. We will assume that there i$ as much light going up as
there is going down and that there is the same amount going right as there is going left.
A surface will therefore receive the same amount of diffuse light energy no matter what
its orientation may be.
Now, some of this energy will be absorbed by the surface, while the rest will be
reflected or reemitted. The ratio of the light reflected from the surface to the total in-
coming light is called the coefficient of reflection or the reflectivity. A white surface re-
flects or reemits nearly all of the incoming radiation; hence its coefficient of reflection
is close to 1. A black surface, on the other hand, absorbs most of the incoming light;
hence its reflectivity is near zero. A gray surface would have a reflection coefficient
somewhere in between. Most objects will reflect some colors of light better than other
colors. Suppose that we have an object which has a high coefficient of reflection for
red light and a low value for the other colors. If we illuminate the object with white
light (which has equal amounts of all colors), then the red portion of the light will be
reflected from the object while the other colors will be absorbed. The object will look
red. In fact, this is the mechanism which gives objects their colors. If we are using a
color graphics display which gives each pixel a red, green, and blue intensity value,
we would have to specify three reflectivity values (one for each of the three colors) in
order to know the color and shading properties of the object. But to keep things sim-
ple, we will begin by considering gray-level displays, and a single reflection coeffi-
cient number (R) will tell us what shade of gray our object is painted.
We said that the amount of light striking the object will not change with the ob-
ject’s position because the light is coming uniformly from all directions, but will the in-
tensity of the light reflected to our eye change with different orientations? The answer
is that the object will look equally bright no matter how it is positioned. This result
arises from the cancellation of two effects. The first effect is that more light is emitted
perpendicular to the surface than parallel to it. The precise relation is called Lambert’s
law, and it states that the reflection of light from a perfectly diffusing surface varies as
the cosine of the angle between the normal to the surface and the direction of the re-
flected ray. (See Figure 10-1.)

Normal >

Light

FIGURE 10-1
7 The direction of light is measured from the
Surface surface normal.
 LIGHT, COLOR, AND SHADING 347

Thus the amount of light coming from a point on the surface would decrease as
the cosine of the angle as that surface is rotated away from our viewpoint. The compen-
sating effect arises from the fact that we don’t see points but areas.
As the surface is turned away from our eyes, the number of light-emitting points
within an area of view increases, and this increase happens to be by the reciprocal of
the cosine. Therefore, as the surface is turned away from our eyes, we get less light
from each point; but the points appear to be squeezed closer together, and the net effect
is that the brightness of the surface is unchanged. (See Figure 10-2.)
A similar cancellation of effects occurs as the surface is moved farther from the
viewpoint. Light coming from the surface is then spread out over a larger area. The
area increases by the square of the distance, so the amount of the light reaching the eye
is decreased by this same factor. By the same arguments, however, the objects being
viewed will appear to be smaller, again by the same factor. So, while we have less
light, it is applied to a smalier area on the retina and the brightness of the surface re-
mains unchanged. (See Figure 10-3.)
We can now give an expression for the brightness of an object illuminated by dif-
fuse background light. If B is the intensity of the background light and R is the object’s
reflectivity, then the intensity of light coming from any visible surface will be
v = BR (10.1)
If we allow the user to set the values of B and R, he can create light or dark scenes and
‘‘paint’’ his objects different shades of gray. But in this simple model, every plane on a
particular object will be shaded equally. This is not, however, the way real objects

FIGURE 10-2
Viewing the surface at an angle increases the light-producing
area in view.
348 CHAPTERTEN

ea
The amount of light decreases with distance

The apparent size of objects decreases FIGURE 10-3 r

with distance The canceling effects for changes of distance.

look; and for a more realistic shading model, we must also include point sources of il-
lumination.

POINT-SOURCE ILLUMINATION
Point sources are abstractions of real-world sources of light such as light bulbs, candles,
or the sun. The light originates at a particular place, it comes from a particular direc-
tion over a particular distance. For point sources, the position and orientation of the ob-
ject’s surface relative to the light source will determine how much light the surface will
receive and, in turn, how bright it will appear. Surfaces facing toward and positioned
near the light source will receive more light than those facing away from or far re-
moved from the illumination. Let’s begin by considering the effects of orientation. By
arguments similar to those we made earlier, we can see that as a face is turned away
from the light source, the surface area on which a fixed-sized beam of light falls in-
creases. This means that there is less light for each surface point. The surface is less
brightly illuminated. (See Figure 10-4.)
The illumination is decreased by a factor of cos I, where I is the angle between
the direction of the light and the direction normal to the plane. The angleI is called the
angle of incidence. (See Figure 10-5.)
If we have a vector of length 1 called L pointing toward the light source and a
vector of length 1 called N in the direction normal to the surface, then the vector dot
product gives

cosI = L:N (10.2)

Suppose that P amount of light comes from the point source. Then the shade of a sur-
face of an object will be given by
 LIGHT, COLOR, AND SHADING 349

FIGURE 10-4
A surface facing the light source will receive more light than
one at an angle.

v = BR + PR(L-N) (10.3)
The point-source term has the light from the point source, times the cosine of the angle
of incidence (giving us how much light illuminates each point on the surface), multi-
plied by the coefficient of reflection (giving us how much of this light is reemitted to
reach the viewer).

SPECULAR REFLECTION
Our model can be improved even more. Light can be reflected from an object in two
ways: there is a diffuse reflection, which we have included above, and there is specular
reflection. The diffuse reflection depends only upon the angle of incidence, and the re-

Normal

Light

FIGURE 10-5
Y, The angle of incidence.
350 cCHAPTERTEN

flected light can be colored, since the coefficient of reflection is involved. Specular re-
flection acts quite differently. It is the type of reflection which occurs at the surface of
i nearly a single
a mirror. (Note the reflected light in Plates 5 and 6.) Light is reflected in
direction (not spread out over all directions in accordance with Lambert’s law). Plastics
and many metals have a specular reflection which is independent of color; all colors
are equally reflected. We shall assume this is the case for our model.
For specular reflection, light comes in, strikes the surface, and bounces right
back off. The angle that the reflected beam makes with the surface normal is called the
angle of reflection O, and is equal in magnitude to the angle of incidence. (See Figure
10-6.)
The model for specular reflection which we shall present was proposed by Blinn
[BLI77]. According to this model, there are four factors which contribute to specular
reflection. They are the distribution of the surface facets, the direction of the surface,
the absorption of the surface, and the blockage of light due to the surface roughness.
The model assumes that the surface is not perfectly smooth, but rather is made of
tiny mirrorlike facets. Most of the facets will face about the same direction as the over-
all surface, but some will be alittle different and will reflect some of the light in differ-
ent directions. How glossy an object is depends on how much variation there is in the
surface facets. A very glossy object has almost all facets aligned, and light is reflected
in a sharp ray. An object with a broader distribution of facet directions will spread out
the light; reflections will blur and merge; the object’s surface will appear flat. A func-
tion which describes the distribution of facets is given by

Deck ee

The vector H is a unit vector in the direction halfway between the light and the eye.
The light from the specular reflection can be seen when the direction from the object
to the eye is the same direction as that of the reflected light. Another way to state this is
to form a vector halfway between the direction of the incident light and the direction to
the eye. (See Figure 10-7.)

Reflection Light

FIGURE 10-6
Specular reflection.
 LIGHT, COLOR. AND SHADING 35]

Halfway Light
between source
H

FIGURE 10-7
Defining the vector halfway between the light source and the viewer’s eye.

Compare this vector to the surface normal, which is halfway between the inci-
dent and reflected rays. (See Figure 10-8.) If the directions match, then the eye can see
the reflected rays. Let’s first see how to compute the vector H. If L is a vector of length
1 in the direction out of the surface along the incident light ray, and E is a vector of
length | in the direction of the eye, then

t br £
= ———_ 10.5
|L+E| I)

will be a vector of length 1 pointing halfway between them. Now we want to show that
Equation 10.4 gives the desired behavior for light reflected from a surface of facets,
most of which are aligned with the surface normal. If N is a vector of length | in the
direction of the surface normal, then the dot product N - H will give the cosine of the
angle between the halfway vector and the surface normal. When the angle between

H
Reflection

FIGURE 10-8
Vectors used to determine when the eye can see the reflection.
352 CHAPTERTEN

these two vectors is zero, the eye will see reflections from the facets which are aligned
with the overall surface direction. Since this is the majority of the facets, we expect D
to be largest for this case. At angles near 0, the cosine will be near 1. In Equation 10.4
the term 1 — (N : H) will be close to 0, which causes the dehominator to be small, and
D is its strongest for this direction. But the denominator can never be smaller than k,
and D cannot become larger than |. Here k is a positive parameter which describes the
glossiness of the material. For k close to 0, the material is glossy; as the eye moves
away from the angle of reflection, the 1 — (N - H) term moves away from 0 and
quickly overwhelms k. So for small k, we see that D will be small unless N is close to
H, at which point the expression approaches |. We will only be able to see light if we
are looking along the angle of reflection. But for large k, the surface is uniform or flat.
D will always be close to 1, so we will see some of the light no matter which direction
we look. (See Figure 10-9.)

Small k: Glossy

Large k: Flat
FIGURE 10-9
The effect of k.
 LIGHT, COLOR, AND SHADING 353

A second factor is how much of the lighted surface is turned away from our eyes.
As we have seen, when a surface is turned away from our eyes, the number of light-
emitting points increases by a factor of the reciprocal of the cosine of the angle [i.e.,
1 / (N- E) J. For the diffuse case, this effect was canceled by Lambert’s law, but for
specular reflection we should retain it.
The third factor is that not all of the light is reflected; some of it is absorbed (or
transmitted if the object is transparent). This depends upon the angle of view. The
theoretical expression for the amount of light reflection for nonabsorbing materials is
given by the Fresnel equation

=
(=) f, bere - 1} (10.6
e+ ey ice sey
where
c=E-H (10.7)
is the cosine of half the angle between the eye and the light, and g is given by
ga Cat Ao ly? (10.8)
Here n is the effective index of refraction for the material. Transparent objects have an
index of refraction near 1, giving small F values; whereas metals have a high value of n
which results in F near 1, meaning that most of the light is reflected.
Finally, the roughness of the surface may block some of the light. (See Figure
10-10.) An expression for the amount of reflected light blocked by the surface rough-
ness is
G, = 2(N-H).(N:: E)/(E: ) (10.9)

The incident light can also be blocked by the surface roughness

G, =,2.(N -.H) (N,: L),,& - H) (10.10)

The light blockage factor G which should be used is

: G = min{[I, min(G,, G,)] (10.11)
Putting these factors together gives the net proportion of the light available for specular
reflection.
DGF
(10.12)
E-N

We can now extend our shading function to include specular reflection.

v = BR + PR(L'N) + PDGF/ (E:N) (10.13)
a
Finally, let’s include the effects of distance. In theory, the illumination from
square of the distance between it and the object
point source should decrease as the
change with
being illuminated. Using this relation, however, seems to give too much
354 CHAPTER TEN

Reflected light intercepted

Ses
FIGURE 10-10
Incident light blocked The effects of surface roughness.

distance. Perhaps this is partly due to the fact that most light sources are much larger
than points. An object illuminated by a very large nearby light source would hardly
show any change in illumination with small changes in distance. In any case, we want
some function which decreases with distance less rapidly than the square. And we may
also wish it to approach some finite limit for very small distances, so that we need not
worry about division by zero for a badly positioned light source. We suggest the fol-
lowing function

eee
(10.14)
C+ U

where Cis a positive constant and U is the distance from the light source to the object.
Our shading expression then becomes

P[R(L: N) + DGF/(E: N)]

v= BR+ (10.15)
CHE YU
We could generalize this to several point light sources if we wished.
P[R(L- N) + D,|GF/(E- N)]
= > ee 10.16
a CU; ( )
 LIGHT, COLOR, AND SHADING 355

SHADING ALGORITHMS
Now let’s consider how to introduce automatic shading into our graphics system. We
begin by creating a means to turn the automatic shading on or off. The flag SHADE-
FLAG can be set by the user to indicate whether or not automatic shading should
occur.

10.1 Algorithm SET-SHADING(ON-OFF) User routine to indicate automatic shading

Argument ON-OFF the user’s shading setting
Global SHADE-FLAG aflag to indicate automatic shading
BEGIN
SHADE-FLAG <— ON-OFF;
RETURN;
END;

The shading formulas which we presented make use of the dot product of vectors
of length 1 for the determination of cosines. We shall therefore often be normalizing
vectors (scaling to length 1). It will be convenient to have a utility procedure to do
this. The following routine normalizes a vector by dividing by its length. The vector’s
original length is also returned.

10.2 Algorithm UNITY(DX, DY, DZ, LENGTH) Utility routine to normalize a vector
and compute its length
- Arguments DX, DY, DZ the vector to be normalized
LENGTH for return of the vector’s original length
BEGIN
LENGTH <— SQRT(DX { 2 + DY t 2 + DZ 7% 2);
DX < DX / LENGTH;
DY <— DY / LENGTH;
DZ <— DZ / LENGTH;
RETURN;
END;

We must determine the shade of each polygon, using vectors and angles between
it and the light sources and viewpoint. These vectors and angles must be determined
before projection (which alters them). We shall alter the DISPLAY-FILE-ENTER
routine to include setting of the CURRENT-FILL-STYLE to the proper shade.

18.3 Algorithm DISPLAY-FILE-ENTER(OP) (Revision of algorithm 8.38) Routine

to enter an instruction into the display file
Argument OP opcode of instruction to be entered
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
PERSPECTIVE-FLAG perspective vs. parallel projection flag
Local X, Y, Z hold the point that is transformed
BEGIN
IF OP < 1 AND OP > —32 THEN PUT-POINT(OP, 0, 0)
ELSE
BEGIN
X < DF-PEN-X;
356 CHAPTERTEN

Y <— DF-PEN-Y;
Z <— DF-PEN-Z;
VIEW-PLANE-TRANSFORM(X, Y, Z);
IF SHADE-FLAG THEN SHADE-CHECK(OP, Xs Y, Z);
IF PERSPECTIVE-FLAG THEN PERSPECTIVE-TRANSFORM(X, Y, Z)
ELSE PARALLEL-TRANSFORM(X, Y, Z);
CLIP(OPREX™ Y,Z):
END;
RETURN;
END;

We have called SHADE-CHECK in order to calculate the appropriate shade for

polygons. The statements in SHADE-CHECK determine the polygon’s normal vector
and its center. The routine then calls MAKE-SHADE to actually find the shading fac-
tor. The SHADE-CHECK procedure is called for every vertex of the polygon, but will
calculate the shade only on the last vertex call. The first call will be for the polygon in-
struction. This causes initializations of the variables and sets a counter for the number
of vertices. For each of the following vertices, we accumulate the average position and
calculate the normal vector as we did for back-face removal. On the last vertex of the
polygon, we complete the center position calculation by averaging the accumulated
vertex coordinates. We then use MAKE-SHADE to determine the appropriate interior
style. Note that the method we use to find the normal vector for the polygon will give
reasonable results when the vertex coordinates are close to but not exactly planar. This
can be useful in finding the shade of small curved surface patches as are discussed in
Chapter 11.

10.4 Algorithm SHADE-CHECK(OP, X, Y, Z) Calculates the shade of polygons

Determines the normal and center; then calls MAKE-SHADE
Arguments OP, X, Y, Za display-file instruction
Global A-SHADE, B-SHADE, C-SHADE for accumulating the polygon normal
X-SHADE, Y-SHADE, Z-SHADE for accumulating the polygon center
X-PREV, Y-PREV, Z-PREV the previous vertex
SIZE-SHADE the polygon size
SIDE-COUNT a counter of polygon sides seen
BEGIN
IF OP > 2 THEN
BEGIN
new polygon instruction
SIZE-SHADE < OP;
A-SHADE < 0;
B-SHADE < 0;
C-SHADE < 0;
X-SHADE < 0;
Y-SHADE < 0;
Z-SHADE < 0;
X-PREV < X;
Y-PREV <—Y;
Z-PREV < Z;
 LIGHT, COLOR, AND SHADING 357

SIDE-COUNT < 0;
RETURN;
END;
IF SIZE-SHADE > 0 THEN
BEGIN
enter another side
A-SHADE < A-SHADE + (Y-PREV — Y) * (Z-PREV + Z):
B-SHADE <— B-SHADE + (Z-PREV — Z) * (X-PREV + X);
C-SHADE < C-SHADE + (X-PREV — X) * (Y-PREV + Y);
X-SHADE <— X-SHADE + X;
Y-SHADE < Y-SHADE + Y;
Z-SHADE < Z-SHADE + Z;
X-PREV < X;
Y-PREV <-Y;
Z-PREV < Z;
SIDE-COUNT <— SIDE-COUNT + 1;
IF SIDE-COUNT = SIZE-SHADE THEN
BEGIN
all vertices seen, so average center position
X-SHADE <— X-SHADE / SIZE-SHADE;
Y-SHADE <— Y-SHADE / SIZE-SHADE;
Z-SHADE < Z-SHADE / SIZE-SHADE;
now determine the shade
MAKE-SHADE;
s SIZE-SHADE < 0;
END;
END;
RETURN;
END;

Now let’s discuss the MAKE-SHADE shading routine. This routine obtains as
global variables the polygon’s normal vector and a center point. It also obtains the po-
sition of the light source and the surface properties of the object. The normal vector
should be scaled to length 1 by the UNITY routine. The difference between the posi-
tiori of the light source and the position of the polygon gives a vector for the direction
to the light. The calculation for the direction to the eye depends upon whether a paral-
lel or perspective projection is used. For a parallel projection, we take the parallel pro-
jection vector. For a perspective projection, the direction to the eye is the difference be-
tween the center of projection and the average polygon position.
After normalizing the light- and eye-direction vectors, they can be added to cal-
culate the vector midway between them. Now the cosines can be determined. The
cosine of the angle of incidence is the dot product of the normal vector and the light
vector. A negative result means that the light is behind the surface and the shade will
result only from the background illumination. We can plug all the parameters into our
shading formula to get a number characterizing the object’s brightness. Here the back-
ground illumination B (named BKGRND in the algorithm) is assumed to have values
between zero and 1, and the point source is given by
P=1-B (10.17)
358 CHAPTERTEN

The result is then scaled by an overall brightness factor BRIGHT. After determining
the brightness of a polygon, the result is passed to the routine INTENSITY, which sets
the appropriate polygon interior style.
.

10.5 Algorithm MAKE-SHADE Routine to calculate and set the shading style for a
polygon
Global XL, YL, ZL the location of the light point source
PERSPECTIVE-FLAG the perspective vs. parallel projection flag
XC, YC, ZC the center of perspective projection
VXP, VYP, VZP the direction of parallel projection
BKGRND the proportion of the light in background illumination
REFL the reflectivity of the polygon
K-GLOSS the polygon shininess
BRIGHT the overall illumination scale vector
X-SHADE, Y-SHADE, Z-SHADE the position of the polygon
A-SHADE, B-SHADE, C-SHADE the normal vector to the polygon
NRF the index of refraction of the surface
Local DXL, DYL, DZL the direction of the light source
U the distance to the lignt source
DXE, DYE, DZE the direction of the eye
DXH, DYH, DZH vector halfway between eye and light
COSNL the cosine of the angle of incidence
COSNH the cosine of angle between normal and halfway vectors
COSNE the cosine of the angle between normal and eye vectors
COSEH the cosine of the angle between eye and halfway vectors
G factor for self-shading due to surface roughness
GF, T intermediate values
F reflection factor from Fresnel’s equation
D facet distribution of the surface
SHADE-SETTING resulting shading value
DUMMY a dummy variable
BEGIN
normalize the surface normal vector
UNITY(A-SHADE, B-SHADE, C-SHADE, DUMMY);
calculate other vectors
DXL <— XL — X-SHADE;
DYL <— YL — Y-SHADE;
DZL <— ZL — Z-SHADE;
UNITY( DXL, DYL, DZL, U)
IF PERSPECTIVE-FLAG THEN
BEGIN
DXE <— XC — X-SHADE;
DYE <— YC: — Y-SHADE;
DZE <— ZC — Z-SHADE;
END
ELSE
BEGIN
DXE <— VXP;
DYE < VYP;
 LIGHT, COLOR. AND SHADING 359

DZE < VZP;

END;
UNITY(DXE, DYE, DZE, DUMMY);
DXH < DXE + DXL,;
DYH < DYE + DYL;
DZH <— DZE + DZL;
UNITY(DXH, DYH, DZH, DUMMY);
calculate the cosines
COSNE < A-SHADE * DXE + B-SHADE « DYE + C-SHADE * DZE;
IF COSNE = 0 THEN RETURN;
it’s a back face
COSNL < A-SHADE * DXL + B-SHADE « DYL + C-SHADE * DZL;
IF COSNL = 0 THEN
BEGIN
does not face the light source
SHADE-SETTING <— MIN(1, BRIGHT « BKGRND * REFL);
SET-FILL-STYLECINTENSITY (SHADE-SETTING));
RETURN;
END;
COSNH < A-SHADE « DXH + B-SHADE « DYH + C-SHADE * DZH;
COSEH < DXH « DXE + DYH * DYE + DZH * DZE;
surface roughness
IF COSNE < COSNL THEN G < 2 « COSNE * COSNH / COSEH
ELSE G < 2 x COSNL *« COSNH / COSEH;
IFG > 1THENG <1;
reflection factor
GF <— SQRT(NRF « NRF + COSEH *« COSEH — 1);
T <((GF + COSEH) « COSEH — 1) /((GF — COSEH) * COSEH + 1);
F<T*T+ 1;
T <— (GF — COSEH) / (GF + COSEH);
FoeFxeT*T/2;
facet distribution
D < K-GLOSS / (K-GLOSS + 1 — COSNH);
D<D+*D;
calculate the shading value
SHADE-SETTING < MIN(1, (BKGRND « REFL + (1 — BKGRND) * (REFL *
COSNL + G*F* D/COSNE)/(1 + U)) * BRIGHT),
SET-FILL-STYLE( INTENSITY(SHADE-SETTING));
RETURN;
END;

The INTENSITY routine converts shading values into settings for the polygon
interior style. The actual implementation of this routine can depend upon the display
device being used. In particular, the number of intensity values actually available and
their corresponding fill-style parameters can differ for different implementations. The
range of SHADE-SETTING values passed to the INTENSITY routine is 0 to 1. If the
display device has possible intensity settings of 1 through N, these values may be gen-
erated by
360 cHAPTERTEN

INT(SHADE-SETTING * (N — 1) + 1)
This is a linear transformation which assumes that the intensity varies linearly between
settings. This is not true for some display devices, and in.such cases, different for-
mulas which compensate for the nonlinear hardware may have to be used. This is dis-
cussed further in the section on gamma correction. If the system has been written such
that fill style 1 is the darkest, style 2 is lighter, and so on up through style N, then the
above formula will give the correct style settings. An example of a case where the in-
tensities may not correspond simply to the interior style numbers is the case where out-
put is done on aline printer and different characters are used to get the different inten-
sities. When the relation of style number to intensity is complex, an array can be used
for the conversion. We store the fill-style codes in the array. The code for the darkest
style is saved in the first array cell, the next darkest in the second cell, and so on, so
that the code for the lightest style is in the Nth cell. Then we use the above formula to
determine which cell to examine, and we retrieve from that cell the correct style set-
ting. To actually make the change of style, we put the new style instruction into the dis-
play file. An example of an INTENSITY routine which could be used with a line
printer is given below.

10.6 Algorithm INTENSITY(SHADE-SETTING) Example of routine to enter correct

interior style for shading
Argument SHADE-SETTING the polygon’s shading value
Global SHADE-STYLES an array of 16 interior styles for shading
Local SHADE-INDEX index of the style setting for the shade
BEGIN
SHADE-INDEX < INT(15 * SHADE-SETTING + 1);
INTENSITY <- SHADE-STYLE[SHADE-INDEX];
RETURN;
END;

We have used a number of parameters to characterize the light source and the ob-
ject being illuminated. Of course, we must provide the means for the user to set the
values for these parameters. We shall therefore write routines for placing the user’s
specifications into global variables. Objects will be characterized by their reflectivity
(REFL), their index of refraction (NFR), and their glossiness (K-nooo y.The follow-
ing routine sets the object parameters.

10.7 Algorithm SET-OBJECT-SHADE(REFLECTIVITY, REF RACTION,

GLOSS) User routine to set an object’s shading parameters
Arguments REFLECTIVITY object’s coefficient of reflection
REFRACTION the index of refraction
GLOSS the narrowness of the reflected rays
Global REFL, NRF, K-GLOSS global storage for the object’s parameters
BEGIN
REFL < REFLECTIVITY;
NRF < REFRACTION;
 LIGHT, COLOR, ANDSHADING 361

K-GLOSS <— GLOSS;

RETURN;
END;

We have allowed the object parameters to be changed at any time, so that as the
scene is constructed, different objects may be given different appearances. We wish,
however, to treat the lighting parameters somewhat differently. The lighting arrange-
ment will be considered part of the viewing specification, and like the other viewing
parameters discussed in Chapter 8, the lighting parameters should be fixed throughout
the display-file segment. To do this, we shall have two sets of parameters: one set
which the user can alter at any time, and a second set which is given the current user’s
specification whenever a display-file segment is created. It is the second set which is
actually used in the shading calculation. The following routine gets the user’s specifica-
tion.

10.8 Algorithm SET-LIGHT(X, Y, Z, BRIGHTNESS, BACKGROUND) User

routine for saving the lighting parameters
Arguments X, Y, Z the location of the light source
BRIGHTNESS an overall-intensity scaling factor
BACKGROUND the portion of the light which is indirect
Global XL1, YL1, ZL1, BRT, BKG storage for the user’s specification
BEGIN
Xe
DY Mee Ys
eee
BRT <— BRIGHTNESS;
BKG <— MAX(0, MIN(1, BACKGROUND));
RETURN;
END;

When a display-file segment is created, we shall copy the current lighting

parameters into the global variables which will be used in the shading calculation. We
shall also convert the light-source position to the view plane coordinate system. The
COPY-LIGHT routine carries out these operations.

10.9 Algorithm COPY-LIGHT Routine to update the current lighting parameters

Global XL1, YL1, ZL1 the current light-source position
BRT the current brightness
BKG the current background illumination
XL, YL ZL lighting position for the display-file segment
BRIGHT the brightness for the display-file segment
BKGRND the background illumination for the display-file segment
BEGIN
BRIGHT < BRT,
BKGRND < BKG;
XESS
WE ay LI;
362 CHAPTERTEN

ZL <— ZL1;
VIEW-PLANE-TRANSFORM(XL, YL, ZL);
RETURN;
END; .

The COPY-LIGHT routine should be called as part of the establishment of the

viewing parameters. We therefore modify the NEW-VIEW-3 routine as follows:

10.10 Algorithm NEW-VIEW-3 (Revision of algorithm 8.39) Create a new overall

viewing transformation
BEGIN
MAKE-VIEW-PLANE-TRANSFORMATION;
MAKE-Z-CLIP-PLANES;
NEW-VIEW-2;
COPY-LIGHT:
RETURN;
END;

To complete our shading algorithm, we must include initializations for the

parameters which we have created.

10.11 Algorithm INITIALIZE-10 Initialization of the shading parameters

BEGIN
SIZE-SHADE < 0;
SET-LIGHT(O, 0, 1, 1, 0);
INITIALIZE-9;
SET-OBJECT-SHADE(1, 10, 10)
SET-SHADING(FALSE);
RETURN;
END;

The light source is initialized to location (0, 0, 1) with brightness 1. Brightness

values will normally be about | to 2, but may be larger to compensate for a distant
light source or low reflectivity of an object. The background illumination is initialized
to 0. This parameter should range between 0 (no indirect lighting) and 1 (totally indi-
rect illumination). Objects are initialized as totally reflective but not very glossy.

SMOOTH SHADING OF SURFACE

APPROXIMATIONS
So far we have dealt only with flat polygons. If we wanted to draw a curved shape (a
doughnut, ball, can, cup, or airplane), we could approximate it with many small poly-
gons. (See Figure 10-11.)
If we shade the approximate surface, we can often see the individual polygon
edges. The edges are enhanced by the properties of our eyes to produce an effect called
Mach banding. When a uniform dark surface meets a uniform light surface, the dark
surface appears even darker along the edge, and the light surface even lighter. This
makes the edge stand out. (See Figure 10-12.)
 NNNE
xRN NNEC
IN
x bhehhhheices
RS
ae
cacunneneeneeaetese™ SEEEEEEEE

Sk
cuauusancensexeens
eran
ee Suceaaece
eeeoo sis Sioa

aes
2%Ridcerehets
‘3 Bt <3x3‘3 % oe 3 Se:m33 +% % 3se %i %it %% %oe %x ie:5c

xe
aac preees Seeder

Rau
Kia
nea: sasoe ranean
we nnnnnn
ionnaBr
Bacau
RMR
RAE

ux
KAKI

De

nae
I SE rs
BSI
2 rerperrr perrprrpr
eee
ee ee
KK aKAR Sa MARR
Sen

Soebea EX
HK eRe RRRMK REARMHERR eprrrrrprrprrerreyrerrrprerrs

DORKS
RANE aRA
ORES RSS
EA RS BESASE

mation
polygons.

noticea
FIGURE 10-12
Approx Imating
FIGURE 10-11

iceable.
a curve
w

363
Flat shading makes the approxi-
ith flat
364 CHAPTER TEN

A method was suggested by Henri Gouraud for varying the shade across polygon
surfaces so that the shade at a polygon’s edge matches that of its neighbor. (See Figure
10-13.) To do this, we need more information than just asingle shade value per poly-
gon. Instead, we need a shade value at each vertex.
We could, for example, extend our polygon routines to enter an intensity value at
each vertex, along with the position. This information would have to be included in the
display file. The rest of the work would be done by the polygon-filling routines. We
use the same method that is used to determine depth values for every polygon point for
use in a Z buffer. Recall that our polygon-filling method moves down the polygon a
scan line at a time, calculating the x edge values and setting the pixels between them.
The new x values could be found by subtracting the inverse slope from the current
values. In a similar way we can arrive at a shade value for the edge at each scan line.
We start with s, at y,, and for each step down in y, we change the shade by (s, — s,) /
(y2 — y,). Now for each scan line we have bounding x values (x, to x,) and bound-
ing shade values (s, to s,). We modify our FILLIN algorithm to change pixel inten-
sities smoothly between these limits. At x,, shade s, is used. At x, + 1, the shade s, +
[(s, — 8,)/ (%, — X,)) is used. AOx, + n, the shade is's, + n [(s, — s)) Ax, = a21-
So shading changes linearly along edges and between edges, yielding a smoother,
more natural appearance.

%a
i%
se‘3
xi
*is
+%
3+
ii
ii
i+4
:
i%
5i
%

FIGURE 10-13
Gouraud shading.
 LIGHT, COLOR, AND SHADING 365

Specifying the shade of each vertex in the polygon routine leaves it up to the user
to determine the appropriate shading. It is also possible to perform automatic shading
of curved surfaces using Gouraud’s interpolation scheme. The problem is to automati-
cally determine the shade of the surface at each vertex. The formulas that we have pre-
sented may be used, only the shade value at a vertex should be based on the direction
of the true surface at that point and not the direction of the polygon approximation.
(See Figure 10-14.)

TRANSPARENCY
Our shading model has not considered transparent objects. The surface of a transparent
object may receive light from behind as well as from in front. Such objects have a
transparency coefficient T as well as values for reflectivity and specular reflection. The
transparency coefficient may depend upon the thickness of the object. If one layer of
tinted glass lets through only half of the light, then two layers will let through only one
quarter. The transmission of light depends exponentially on the distance which the
light ray must travel within the object.
hesntes (10.18)
In this expression, (d) is the distance the light must travel in the object. The coeffi-
cients (t) and (a) are properties of the material. The (t) determines how much of the
light is transmitted at the surface (instead of reflected), and (a) tells how quickly the
material absorbs or attenuates the light. (See Figure 10-15.)
If we view a light source through a transparent object, the properties of the sur-
face (t) may be modeled in much the same way as for a reflecting surface. We are
mainly interested in how much light passes through the surface. This is the light which
is not reflected and is given by the Fresnel factor (1 — F). For most objects, the light
both enters and exits, and a factor is needed for both surfaces.
The transparency and absorption coefficients can depend on color. Some objects
let only red light through, while others attenuate the red and blue, letting only green
pass. If we are dealing with colored objects, we need three pairs of transparency and
absorption coefficients. For very clear objects, we can neglect the attenuation with dis-
tance or include it as an average value as part of (t).
A proper treatment of transparent objects should also consider the fact that light
changes direction when it crosses the boundary between two media. (See Figure
10-16.) This effect is called refraction. The change in direction is related to a property
of materials called the index of refraction (n). The exact relation is called Snell’s law.

FIGURE 10-14
Use the direction of the true curve at each
vertex of the approximation.
366 CHAPTERTEN

FIGURE 10-15
The absorption of light by an object de-
pends on the length of the optical path
within the object.

sin®, n,
— = — (10.19)
sin®, n,
The light ray is bent both when it enters a transparent object and when it leaves. To cor-
rectly show transparent objects, then, we have to retain back faces so that we can cal-
culate the light path in and out of objects and also determine the length of the path
within the object. This would be difficult to add to our system; however, there is a sim-
ple extension which can yield transparency effects.
If we assume that (1) for a given surface, the transparency coefficient for the ob-
ject is a constant, (2) refraction effects can be ignored, and (3) no light source can be
seen directly through the object, then the light coming through the object is just the
transparency coefficient times the light coming from objects behind it. Our painter’s al-
gorithm overwrites objects in the back of the scene with the objects in the front. In ef-
fect, the light from the back object is replaced by that from the front object. But we
can make the front object appear transparent if instead of blindly replacing old shading
values, we multiply them by the transparency coefficient and add them to the values
for the new surface. The shade actually displayed will correspond to light from the new
object’s surface plus the light behind, which is transmitted through the object.

Wee per atv (10.20)

®, FIGURE 10-16
Refraction.
 LIGHT, COLOR, AND SHADING 367

Here (v) is the total amount of light, (v,) is the amount reflected from the surface as is
given in Equation 10.16, (t) is the transparency coefficient, and (v,) is the light coming
from behind the object, which may be determined from the frame buffer value. We will
require an inverse function for INTENSITY so that given the value in the frame buffer,
we can determine the original SHADE-SETTING of the background object for use in
the transparency calculation. Transparent objects often have a low reflectivity, and the
reflected light (v,) comes from specular reflection.
This simple approximation does not always yield realistic images, particularly
where curved surfaces are being constructed. To improve it, we would like to include
the angular behavior of the Fresnel reflection vs. transmission at the surface and also
the attenuation due to thickness. Imagine light coming from some background object,
passing through a transparent object, and reaching the eye. (See Figure 10-17.) If the
transparent object has parallel sides (as do glasses, bottles, and windows), then the dis-
tance the light must travel through the object depends on the angle at which we view it.
When viewed straight on, the distance the light travels within the object is less than
when viewed at a glancing angle. Also, when the object’s surface is perpendicular to
the path of the light, the maximum amount is transmitted (instead of reflected). A sim-
ple approximation for this behavior is
ipa rt) Ne Bei ba (10.21)
This is the cosine of the angle between the eye and the surface normal raised to a
power. The cosine is strongest when we view the surface straight on and drops off for
glancing views. The power just enhances the effect. Values of a of 2 or 3 give reason-
able effects.

REFLECTIONS
We have seen how to determine the effects of light coming directly from illumination
sources, but in a realistic scene, every object can act as a light source. We should be
able to see the specular reflections of objects, not just of lights. We look at a point p,
on an object and ask what light is coming from that point and from where does it origi-
nate. We have seen how to find the light coming from diffuse illumination and directly
from point sources. We can also follow a reflected ray back from our eye to the point
and determine the incident ray. If we move back along the incident ray and reach a
point p> on some other surface, then we know that the light (I) from this second point
will be reflected at p,. (See Figure 10-18.)

FIGURE 10-17
The reduction in light from reflection and ab-
sorption depends on the orientation.
368 CHAPTERTEN

* .

P,

FIGURE 10-18
Y Light reflected from another object.

This will add a term to Equation 10.16 for the contribution of p, of the form
IF'/(E - N)
(10.22)
Carew
where F’ is calculated according to Equation 10.6 just as F, onlyc used in this calcula-
tion is given by
¢ = E> N (10.23)

The light (1) for the point p, should be calculated as if the eye is located at p,. For such
calculations, all surfaces must be considered, even back faces (because a batk face
may be visible in a reflection)‘ The method is discussed further in the section on ray
tracing.

SHADOWS
The problem of shadows is very similar to the hidden-surface problem. A shadowed
object is one which is hidden from the light source. Shadow and hidden-surface calcu-
lations are often performed together for efficient computation.
One approach to the shadow problem is to repeat the hidden-surface calculation
using the light source’as the viewpoint. This second hidden-surface calculation divides
the polygons into shadowed and unshadowed groups. Surfaces which are visible and
which are also visible from the light source are shown with both the background illumi-
nation and the light-source illumination. Surfaces which are visiblé but which are hid-
den from the light source are displayed with only the background illumination.
Another approach to the shadow problem is the use of shadow volumes. Consider
a polygon and a light source. There is a volume of space hidden from the light by the
polygon. This is the polygon’s shadow volume. We can construct this volume from the
polygon itself and from the rays from the light which touch the polygon edges. (See
Figure 10-19.)
We can compare all other visible polygons to this volume. Portions which lie in-
side of the volume are shadowed. Their intensity calculation should not include a term
from this light source. Polygons which lie outside the shadow volume are not shaded
 LIGHT, COLOR, AND SHADING 369

FIGURE 10-19
Shadow volume.

by this polygon, but might be shaded by some other polygon so they must still be
checked against the other shadow volumes.
For simplicity, assume that all polygons are convex. (There are techniques for
breaking up any polygon into convex components. In particular, polygons can be de-
composed into triangles or trapezoids, which are always convex.) The shadow volume
for a convex polygon is a convex object, which means that we can use the generalized
clipping techniques to find shadowed areas. Points within the shadow volume are
shadowed, and a point lies within the volume if it lies on the interior side of the poly-
gon plane and all the shadow planes. So shadowing can be done by ‘‘clipping’’ all vis-
ible polygons to all possible shadow volumes.
The shadow volumes may be included as part of the hidden-surface processing.
We include the surfaces of the shadow volumes along with the object surfaces. We in-
clude both front and back faces for shadow volumes, but tag them as such. We sort all
polygons to determine their visibility. Shadow surfaces are invisible, and are not drawn
but are counted. Where the number of front shadow faces in front of the visible poly-
gon exceeds the number of back shadow faces, the polygon is in shadow.
We can use Z-buffer techniques to simplify the sorting. Suppose we have two Z
buffers, one for determining the foremost visible surface and one for finding shadows.
We begin with the Z-buffer hidden-surface removal algorithm described in Chapter 9.
After this step, the visible surface Z buffer will contain the z coordinates of all visible
points. The next step is to determine the shadow volumes for the scene. We can
examine the surfaces in these shadow volumes to find which are the front and back
faces. For each shadow volume, we can enter the z coordinate of the back-face pixels
in the shadow Z buffer. We can also determine the z values for the pixels of the front
faces on the shadow volume. Now if the z value of the front face of the shadow (which
we just calculated) is in front of the visible surface (the z value in the visible surface Z
buffer) and this in turn is in front of the back of the shadow volume (the z value in the
shadow Z buffer), then the visible surface lies within the shadow volume. (See Figure
10-20.)
Shadows from point sources of light are sharp and harsh. In the real world we
seldom find light coming from points. The sun and moon are disks; artificial light
comes in a variety of shapes. Shadows from these finite shapes have softer edges.
370 CHAPTER TEN

Back face of shadow volume

from the shadow Z buffer

Object’s front face

from the visible
surface Z buffer

Z
Front face of shadow volume
FIGURE 10-20
Using a shadow Z buffer to find shaded objects.

Much of the current work on shadows is in modeling these finite light sources to give
more realistic results.

RAY TRACING
There is a simple brute-force technique which can yield startlingly realistic computer-
generated images. (See Plate 15.) Its basic form is easy to implement and requires little
of the machinery of the graphics system we have been constructing. It can handle
curved surfaces as well as flat polygons. It includes perspective projection, hidden-sur-
face removal, reflections, shadows, and transparency. The technique is called ray trac-
ing, and the problem with it is that it is notoriously slow. The idea is to determine the
intensity for each pixel one at a time by following a ray back from the viewpoint
through the pixel and into the object space. We can compute the intersection of that ray
with all the surfaces. If none is encountered, then the pixel has the background shade.
If one or more surfaces are impacted by the ray, we find the one closest to the view-
point. It is the one which will be seen (the rest are hidden by it). This, in effect, solves
the hidden-surface problem for every pixel. (See Figure 10-21.)

FIGURE 10-21
Ray tracing.
 EC

Use of a menu for interaction with the

program. (Courtesy D.N.J. White,
Chemistry Department, Glasgow
University, Scotland Dr. White's figures
were also published in JEEE Computer
Graphics and Applications, vol. 5, no. 10,
a pp. 8-14, 1985. © 1985 IEEE, reprinted
by permission.)

Plate 2 ry =
VON |Chet 2 |
This figure uses front and back clipping Ai Cd

to display. a slice of a molecule. (Courtesy i

D.N:J. White, Chemistry Department,
Glasgow University, Scotland. © 1985 IEEE,
reprinted by permission.) |-
mat = es
i
Pry ay OL i TT ‘|
a! Ed
sere
3 en r
ae Be) Bd
Per art wd; serhempi fener Pt aE
: oe as cee]
retrace Pra Ce LU by

Plate 3

The viewing transformation for this mole-

cule can be interactively set by a series
of valuators which are simulated by a
locator and scales on the display. (Courtesy
D.N.J. White, Chemistry Department.
Glasgow University, Scotland. © 1985 IEEE.
ae tM meean (OLD)

Po
{ }
i
CC ee eo
Pra | omer
iacee eea we we
TFET sea
: fey mr

Plate 4

A molecule composed of atoms displayed

as spheres with hidden-surface removal.
(Courtesy D.N.J. White, Chemistry
Department, Glasgow University, Scotland.
© 1985 IEEE, reprinted by permission.)
Plate 5
aca 5).LN Lelom PEE ROUT Rseem COU meer uale
See OMOOM OURO
Designer. (Courtesy of Abel Image Research, Los Angeles, California.
Randy Roberts Director/
© 1985 Abel Image. Research.)

Plate 6

Knight from the opening sequence of -

PLT ARY a OME UT OO ae
motion is accomplished by analyzing the
actions of a live model. Randy Roberts
Director/Designer. (Courtesy of Abel Image
"Research, Los Angeles, California.
© 1985 Abel Image Research.)
 ENA

An image from “High Fidelity”

of Ava dancing with her
umbrellas. Ava’s texture was
digitized as a flat 2D image and
then wrapped around her 3D
geometric shapes. Randy Roberts
Director/Designer. (Courtesy of
Abel Image Research,
Los Angeles, California.
© 1985 Abel Image Research.)

Plate 8
Andre’s Forest. The trees, grass, and flowers in this scene were stochastically generated using particle
systems. A specialized shading function was used to cast sunlight on the trees and to simulate self-
shadowing. The shadows on the grass were generated using a shadow mask derived from the trees.
Created by William Reeves, John Lasseter, and Eben Ostby. (Courtesy Lucasfilm, Ltd., San Rafael,
California. The techniques used to create Andre’s forest are discussed in Computer Graphics,
vol. 19, no. 3, pp. 313-322, 1985 © 1985 Lucasfilm. Ltd.)
bg Econ

A scene from “The Adventures of Andre & Wally B.” This film Reet full classic character
animation, motion blur of the characters, and very complex 3D background sets. (Courtesy
Alvy Ray
Smith, Lucasfilm Ltd., San Rafael, California. © 1985 Lucasfilm, Ltd.)

Plate 10
“Happy Drinking Birds” by
Richard Cohen (Courtesy Pacific ©
Data Images, Sunnyvale,
California. © 1985 Pacific Data
-Images.)
ETCom Ol

“Chromosaurus” by Don Venhaus. (Courtesy Pacific Data Images, Sunnyvale, California.

© 1985 Pacific Data Images.)

Plate 12

“Dishwasher” from a Kitchen Aide

commercial by S. Cohen and S. Legensky.
(Courtesy Intelligent Light, Fair Lawn,
New Jersey. © 1984 Intelligent Light.)
Piate 13

“Backhoe.” a frame from a Tenneco Renae LY Uaioe D. Stipe, S. Legensky. (Courtesy Intelligent
Light, Fair Lawn. New Jersey. © 1984 Intelligent Light.)

Plate 14

(erat say Intelligent Light,

Fair Lawn, New Jersey.
© 1986 Intelligent Light.)
Plate 15

The scene was produced bya ray tracing with cones. This gives more fidelity, better antialiasing, fuzzy
shadows, and dull reflections. (Courtesy John Amanatides, University of Toronto, Canada. The creation of
this image is discussed in Computer Graphics, vol 18, no. 3, pp. 129-135, 1984. © 1984, Association for
Computing Machinery, Inc., reprinted by permission.)

IgEom

The 1931 CIE chromaticity diagram.

(Courtesy Photo Research, Burbank,
California.)

G0
4 44%
2
ul u ui
a
; t
es
| |
 LIGHT, COLOR, AND SHADING 371

Next we can find the illumination of the surface point by calculations such as
Equation 10.16. We may, however, as part of this calculation, look for the intersection
of the rays between the surface point and the light sources with other objects in the
scene. If a ray is cut by another surface, then the corresponding light source is
shadowed for the pixel. This solves the shadow problem for every pixel. The same ray-
following routines that were used to find the surface being drawn can be used to find
shadows, only the starting point and direction of the ray are different. (See Figure
10-22.)
We must add to the intensity value for the surface point the amount of light re-
flected from other objects and, if the point belongs to a transparent object, the light
transmitted through it. The reflected light is equal to the coefficient of specular reflec-
tion times the amount of light coming in along an incident ray. We find the light along
this incident ray by recursively applying the method which we are describing. We
imagine the surface point as the viewpoint and ask how much light comes to it along a
given direction. We examine this new ray for intersections with objects and find the
light from the closest. This may entail further recursive calls, until either the ray makes
no intersections or the accumulated spectral reflection coefficients have reduced the
light to where it is no longer significant. (See Figure 10-23.)
If objects are transparent, then we must also calculate and add in the transmitted
light. As with the reflection case, we calculate the refracted ray which would give rise
to the light along the ray we followed. We recursively apply our method to this ray to
determine the light coming along it. We multiply the amount of light by the transpar-
ency and attenuation factors and add it to the total. (See Figure 10-24.)
Most of the time spent in ray tracing is in finding the intersection of a ray with
the nearest surface. As the complexity of the image goes up, the number of surfaces
which must be checked against the ray also increases. There is much interest and re-
search into ray-tracing techniques, both in modeling the properties of objects and light

FIGURE 10-22
Ray tracing to find shadows.
372 CHAPTERTEN

FIGURE 10-23
Ray tracing to find reflections.

sources to make them more realistic and in finding data structures and algorithms
which make it more efficient.

HALFTONES
Many display devices allow only two imaging states per pixel (on or off, black or
white). This is fine for line drawings, but when displaying surfaces, we would often
like more variety than just black or white. There would be little point to the shading
calculations we have discussed if there were no way to produce gray images. We dis-
cussed in Chapter 3 how our polygon-filling algorithm could be extended to display
patterns. Using such patterns allows us to give different polygons different appear-

FIGURE 10-24
Repeated ray tracing.
 LIGHT, COLOR, AND SHADING 373

ances, and with a judicious choice of patterns, this mechanism can also provide gray
levels for use in shading. But for realistic images, as might be generated by ray tracing
or might be input from a television camera or scanner, we must make further exten-
sions so as to be able to control the apparent intensity in finer detail than we did for en-
tire polygons. We have as a guide the work done by the printing industry. Photographs
are routinely reproduced in books and newspapers using only two states (ink or no
ink). The technique is called halftoning. A continuous tone image is shown through a
screen. This produces a regularly spaced pattern of small dots. The dots vary in size ac-
cording to the original image intensity; so where the image is light, the dots are small
and the background color of the paper predominates. Where the image is dark, the dots
are large and merge together to cover the paper with ink. (See Figure 10-25.)
How can we produce halftones? Our first try might be to divide our binary pixels
into groups corresponding to the halftone cells, and call each group asingle gray-level
pixel. We could develop a set of dot patterns for the groups, making dot patterns for
the group correspond to intensity values for the gray-level pixel. Every time we set a
gray-level pixel to an intensity value, we would really be copying a binary dot pattern
into the group of binary pixels which form it. For example, we could make each 2 Xx 2
pixel square our gray-level pixel. This would give us five possible intensity settings.
We could design a pattern for each setting. (See Figure 10-26.)
Our effective resolution would be half of the binary resolution of the device, but
we would have five intensity values instead of just two. (See Figure 10-27.)
The scheme we have outlined will work, but we can do better. We can produce
images with a better effective resolution and still have as many gray levels. The reason
this can be done is that there exists more than a single possible pattern for many of the
intensity levels. Consider the intensity level of two active pixels for the 2 x 2 halftone
cell. There are six possible patterns. (See Figure 10-28.)
The idea is to select patterns which closely match the structure of the image. For
example, if we are looking at a polygon edge which crosses the halftone cell, as in Fig-

FIGURE 10-25
Enlarged halftone image of a bird in flight.
374 CHAPTERTEN

FIGURE 10-26
A grid of halftone cells.

ure 10-29, then for a dark polygon on a light background the pattern with the right two
pixels shaded would be most appropriate. For a lighter polygon, perhaps only one of
the two right pixels would be colored, but the left pixels should remain light. A simple
method which comes close to implementing this kind of halftoning is an ordered dither.
It is done by constructing a pattern of threshold values, called a dither matrix, which
correspond to the individual pixels in the halftone cell. The pattern can be replicated to
cover the screen so that every binary pixel has an associated threshold value. We then
determine the intensity value for every pixel (full resolution, not just gray-level
pixels). We compare the intensity of each pixel against its threshold value to determine
whether the pixel is turned on or off.
This method also has the advantage of allowing intensity values to range over
any scale of values desired. An image could be constructed in which each pixel has
one of 256 possible intensity values. The image could be shown on a display using a
halftone screen with 5 gray levels, or one with 50 gray levels, or one with 256 gray
levels by just designing the corresponding cell of threshold values. For example, sup-
pose we display such pixels using the five-level 2 X 2 cell. A possible dither matrix is
shown in Figure 10-30. With this pattern, the lower right pixel would be turned on if it
were set to an intensity greater than 50. The upper left pixel would be turned on only if
it were set to an intensity greater than 200.
Notice that we can alter the appearance of the image by changing the threshold
values. The intensity at the display need not correspond linearly to that of the source
image. This is useful when the source image originates from real life, such as a
scanned photograph. Altering the thresholds can adjust for under or over exposure and
can be used to enhance the image. The thresholds also can be used to compensate for
nonlinearities in either the scanning, the display hardware, or our eyes. Our eyes are in
fact sensitive to intensity ratios rather than absolute intensities; so to achieve a set of
gray levels which look equally spaced, we might select threshold Values which change
exponentially (1, 2, 4, 8, 16) or cubically (1, 9, 27, 64).
Implementation of this halftoning scheme is very similar to the method described
in Chapter 3 for filling a polygon with a pattern. In that scheme we used the polygon

FIGURE 10-27
Intensity settings for the halftone cell.
 LIGHT, COLOR, AND SHADING 375

ie
-
FIGURE 10-28
Possible patterns for two active pixels.

style to select a pattern and set the frame buffer pixel to the corresponding pixel ele-
ment. For halftoning, when filling a polygon, we look up the threshold value for each
pixel and compare it with the intensity value for the polygon. The results of the com-
parison are used to decide where to set the pixel. A replacement algorithm for FILLIN
which does this halftoning is as follows:
10.12 Algorithm HALFTONE-FILLIN(X1, X2, Y) Fills in scan line Y from X1 to X2
using halftoning
Arguments X1, X2 end positions of the scan line to be filled
Y the scan line to be filled
Global HALFTONE-CELL array of intensity thresholds
FRAME the two-dimensional (binary) frame buffer array
FILLCHR the intensity of the polygon
Local X for stepping across the scan line
HX, HY for accessing the halftone cell
Constants HALF-X, HALF-Y the x and y dimensions of the halftone cell
ON, OFF the possible pixel values
BEGIN
“ IF X1 = X2 THEN RETURN;
HX <— MOD(X1, HALF-X) + 1;
HY < MOD(Y, HALF-Y) + 1;
FOR X = X1 TO X2 DO
BEGIN
IF FILLCHR > HALFTONE-CELL[HX, HY] THEN FRAMEIX, Y] — ON
ELSE FRAME[X, Y] < OFF;
IF HX = HALF-X THEN HX < 1
ELSE HX <— HX + 1;
END;
RETURN;
END;

Note that the above halftoning scheme and the pattern filling of Chapter 3 can be
combined to allow filling polygons with gray level and photographic style patterns (or

Serer
PELE
FIGURE 10-29
Select patterns to match the image shape.
376 CHAPTER TEN

FIGURE 10-30
_ A dither pattern for the 2 x 2
cell.

Gouraud shading) on a bilevel display. In the combined scheme we would look up both
the threshold value from the halftone description and the intensity value from the pat-
tern or polygon for the given pixel. These would then be compared to determine if the
pixel should be set.

COLOR
Light is an electromagnetic wave and has wave properties. One of these properties is
frequency (v). The frequency describes how many cycles of the wave occur every sec-
ond. A related property is the wavelength (), the distance in space between crests of
the wave. They are related properties because light moves at a constant speed (c = 3 X
10° meters/second in a vacuum). The speed must be the length of a cycle times how
many cycles pass per second, so
c= vA (10.24)
Our eyes are able to detect differences in frequency (or wavelength) for a small range
of the possible values. We can see light with wavelength between about 4 x 107’ me-
ters and 7 X 10~7 meters. Each wavelength appears to us as a color from violet at 400
nanometers(nm) through the rainbow to red at 700 nm. The reason we can see these
colors is that our eyes have three different color sensors which are sensitive to different
parts of the visible spectrum. They are called the blue, green, and red photopigments,
and they give respective peak responses at blue, green, and yellow light, but also give
lesser responses at other values so as to cover the entire visible range. (See Figure
10-31.)
We are not equally sensitive to all colors. Our eyes give the greatest response to
green. We are less sensitive to red and blue light, and cannot see infrared or ultraviolet
light at all. Because of this, the wminosity or perceived brightness.of the light does not
correspond directly to the energy of the light. It takes less energy to increase the bright-
ness of green | lumen (a unit of brightness) than it takes to increase red 1 lumen, which
is in turn less than the energy needed to increase the perceived intensity of blue light
the same amount. (See Figure 10-32.)
Light with a single wavelength produces the sensation of a ‘‘pure’’ color. But
note that there may be other ways to achieve the same response from the eye. For
example, light at 490 nm excites all three photopigments, but the same effect could be
achieved with two less intense lights, one at 450 nm to excite the blue and one at about
540 nm to excite the green and red photopigments. Thus blue-green can be seen either
from a blue-green light or from a mixture of blue light and green light.
 80

70

60

50

40
blue x 20
30

20
photopigments
Relative
eye’s
the
of
sensitivity
10

400 500 600 700

Wavelength (A) nm
FIGURE 10-31
Relative response of the eye’s photopigments.

100

90

80

70

50

40
of
sensitivity
the
Relative
eye
30

20

10 -

600 700
400 500
Wavelength (A) nm

FIGURE 10-32
Total luminosity response of the eye.
377
378 CHAPTER TEN

We can easily mix colors to achieve relative responses from the photopigments
which will not correspond to any of the response patterns for single wavelength
(monochromatic) light. But such patterns can be separated into a piece which looks
like the response from light with equal amounts of all wavelengths and a remainder
which appears like a pure color or its complement. Light with equal amounts of all
wavelengths present looks white to us. Thus mixtures of lights will often appear the
same as some pure color mixed with white. The pure color determines the hue (e.g.,
red), while the amount of white light determines the tint (pink). The more white light
the less saturated is the color. The total amount of light determines its shade or inten-
sity. A diagram was developed by the Commission Internationale L’Eclairage (CIE)
which graphically portrays the eye’s response to colors. (See Plate 16.) On the dia-
gram, every point represents some color. The diagram is designed such that all colors
on a line between two color points may be produced by mixing the light of the end-
point colors. Three points will form a triangle of colors, and all colors within the
triangle may be produced from mixtures of the vertex colors. The colors within the
triangles are called the gamut of colors producible from the colors at the triangle ver-
tices. (See Figure 10-33.) In the center of the diagram is a region corresponding to
white. Along the tongue-shaped boundary are the pure colors of monochromatic light.
Across the base are the purples, which have no monochromatic representative.
The color of light is determined by the relative responses of the three photopig-
ments in the eye. There is also the absolute response which tells us how much total
light there is (how bright it is). In the CIE diagram, just the intensity ratios of the light
are used, so the color is described but not the intensity. This is done as follows: Imag-
ine we have sensors for measuring light. Imagine further that the sensors have re-
sponse curves with peaks at three specially chosen primary colors. Let X, Y, and Z
measure the absolute response of these sensors. Then we can compute the ratios
Xx ¥ Z
SS ————————— Z = ————_— 10.
ROY ABZ : ar Yee ey ee a
It is the x, y values that are shown in the CIE diagram. Note that

Za ke ny (10.26)
so we have only two independent color choices, and they may be shown in a two-di-
mensional diagram. The CIE definition is arranged so that Y is the eye’s total photo-
topic luminosity-function curve, the eye’s total sensitivity to light; which is greatest for
green. The Y value gives the luminance (brightness of the color). The hue which gives
the maximum luminance is green.
Given a chromaticity value (hue and saturation) specified by x and y and an over-
all luminosity Y, we can easily solve for the CIE XYZ coordinates.

X = x (Y/y) and Z = 2z(Y/y) (10.27)

The response of the eye to light and color is not as straightforward as the response of a
physicist’s instruments. The perceived intensity of light coming from an object may
vary greatly with the amount of light from the surrounding area, even though the actual
radiation is the same. The perceived dominant frequency or hue may appear to shift
 LIGHT, COLOR, ANDSHADING 379

0.9

0.8

0.7

0.6

0.4

0.35

0.2

0.1

0
r

FIGURE 10-33
The color gamut for a color monitor.

slightly as white light is added. The perceived hue and saturation may also change with
intensity, dropping out altogether for low radiation levels. In the remainder of the dis-
cussion we shall largely ignore these differences, giving rules for specifying the physi-
cal light and color rather than the psychophysical perception.

COLOR MODELS
There are several other ways to parameterize colored light. These are called color mod-
els. The red-green-blue (RGB) model is often encountered in computer graphics. It cor-
380 CHAPTERTEN

responds to the red, green, and blue intensity settings of a color monitor or television
display. Color television has red, blue, and green phosphors forming a triangle on the
CIE diagram which includes most, but not all, of the possible colors. The RGB color
space may be pictured as rectangular coordinates with black at the origin. Red is mea-
sured on the x axis (R axis), green on the y axis (G axis), and blue on the z axis (B
axis). The grays are on the diagonal line G = R, B = R. The farther from the origin,
the brighter the light. (See Figure 10-34.)
A color display cannot increase its intensity indefinitely; it has some maximum
value. Furthermore, how bright that value appears to the eye depends upon the ambient
room light. It will appear much brighter in a darkened room than it will in bright sun-
light. Because of this absolute intensity, values are not very useful. Instead, we often
normalize the intensity scale so that full intensity corresponds to a value of 1. Then the
set of realizable colors for the device is described by a unit cube.
The XYZ primary colors defined by the CIE are theoretical extensions and not
really the visible red, green, and blue, but we can convert from one coordinate system
to the other with a linear transformation. The transformation depends upon just what is
meant by red, green, and blue. The CIE standard defines

Standard CIE spectral red (700 nm) x = 0.73467, y = 0.26533

Standard CIE spectral green (546.1 nm) xX = 0.27367, y = 0.71741
Standard CIE spectral blue (435.8 nm) x = 0.16658, y = 0.00886
But for the standard television red, green, and blue we have

NTSC red x = 0.670, y = 0.330

NTSC green x = 0.210, y = 0.710
NTSC blue x = 0.140, y = 0.080
And for modern graphics monitors we have

Monitor red x = 0.628, y = 0.346, z = 0.026

Monitor green x = 0.268, y = 0.588, z = 0.144
Monitor blue x = 0.150, y = 0.070, z = 0.780

blue

green

=e FIGURE 10-34
black The RGB color space.
 LIGHT, COLOR, AND SHADING 381

Usually only x and y values are listed because we can find the z values using
Equation 10.26. These values are not enough to tell us how to convert between RGB
and XYZ. We know the colors but not their relative strengths. If we just pick three col-
ors and add them together, we are unlikely to get white; but we want to weight or scale
the primaries we have chosen so that if we do select equal amounts of the red, green,
and blue, we do come out with white. The amount to scale depends on just what we
mean by white, and there are several choices (the color of an electric light, the color of
sunlight at noon, the color of the overcast sky, and others). Monitors are often factory
realigned to the color of a black body heated to 6500 degrees Kelvin (K). This has CIE
values

x = 0.313, y = 0.329 (10.28)

which for luminancy of 1 give
X, = 0.950, Y, = 1.000, Zy = 1.086 (10.29)
So if we were considering the monitor primaries, we would have to supply the scaling
factors r, b, g such that
Xr + Xgg + Xyb

yt + ygg + ypb Yay (10.30)

Z0.+ 2,8 + 2b = Z,
Substituting the actual values gives
0.628r + 0.268g + 0.150b = 0.950
0.346r + 0.588g + 0.070b = 1.000 (10.31)
0.026r + 0.144g + 0.780b = 1.089
We can solve this set of linear equations for r, g, and b. Doing so gives
r = 0.758, g = 1.116, b = 1.165 (10.32)
Multiplying the chromaticities of the primary colors by these weighting factors tells us
how much of the X, Y, Z primaries is contributed by each of the R, G, B primary col-
ors. For example, for the red primary, the amount of X contributed is
X, = (0.628)(0.758)R (10.33)
The total X amount is the sum of the contributions from the R, G, and B components
of a color.
X =x7tR + x,gG + x,bB (10.34)
= (0.628) (0.758)R + (0.268) (1.116)G + (0.150) (1.165)B
=0.476R + 0.299G + 0.175B
Similar statements may be made about Y and Z. We can perform the multiplications
and combine the three equations into a matrix expression. The result is a transforma-
tion for converting from the RGB coordinates of the monitor to XYZ coordinates.
382 CHAPTER TEN

0.476 0.262 0.020

[XY Z]=[RGB]] 0.299 0.656 0.161 (10.35)
0.175 0.082,0.909
Finding the inverse matrix, we can write

2.100. sha lO.. Ul oo

[R"G Bl = EX Y 7) }- 1149.” 2.026. 0.3353 (10.36)
- 0.426 0.033 1.104

Hard-copy devices may use cyan, magenta, and yellow as the primary colors
rather than red, green, and blue. These are the colors of the inks which are placed on
the white paper. They begin with white and use the inks to absorb and remove some
colors. Cyan removes red, magenta absorbs green, and yellow removes the blue. This
complementary relationship between red, green, blue, and cyan, magenta, yellow
makes it easy to convert from one color model to the other. Assuming units where 1
means full color and 0 means no color, and assuming that the cyan, magenta, and yel-
low colors are exact complements of red, green, and blue, then

Ce=_l = R, M=1-G, and Y=1-B (10.37)

or
[CoM )= 1 eG
We can also picture this relation. Start with the unit cube of all colors, with black at a
comer which serves as the origin of the RGB coordinate system. The R, G, and B axes
along the edges of the cube meet at the black corner. The opposite corner on the cube
will correspond to white, and will serve as the origin for the CMY coordinates. The C,
M, and Y axes on the edges of the cube meet at the white corner. Each of the other cor-
ners of the cube can be labeled with one of the primary colors. (See Figure 10-35.)
Printers often use the four colors cyan, magenta, yellow, and black. In theory,
only the three inks cyan, magenta, and yellow (CMY) should be needed. Mixing the
three should produce an ink which absorbs all the light, yielding black; but in practice,
the inks may not absorb completely or mix well, so a fourth black ink is used to set the
shade.
Color television specifies colors by the YIQ color model. The Y is the same as
the Y in the CIE primaries and corresponds to the luminosity or brightness. It is the Y
component which is displayed on a black-and-white television. The I and Q coordi-
nates determine the hue and saturation. The transformation from RGB coordinates is as
follows:

0.60 0.21 0.30

[IQ Y]=[RGB]]|-0.28 -0.52 0.59 (10.38)
70:3009032:.0.44
 LIGHT, COLOR, AND SHADING 383

cyan white

blue

magenta

yellow

black red
FIGURE 10-35
The RGB-CMY color cube.

Human vision is much more sensitive to intensity than it is to color. For this reason, we
do not need to provide as much color information as intensity information. We can
quantize the I and Q coordinates in bigger steps (fewer bits) and give fine detail (more
bits) to the Y coordinate. This allows an overall reduction in the amount of information
which must be sent to achieve a high-quality image. The YIQ model allows a more ef-
ficient representation of a color image intended for a human observer.
Another color model is the hue, saturation (or chroma), and intensity(brightness)
model (HS/). The hue or tone is the pure color of the light. It is the dominant color we
see. Saturation indicates how much white light is mixed with the color (the tint). Sat-
uration of 0 means no color; we see white or gray. The intensity tells how much total
light there is (the shade). Hue, saturation, and intensity are often plotted in cylindrical
coordinates with hue the angle, saturation the radius, and intensity the axis. (See Fig-
ure 10-36.)
The conversion between HSI and RGB models can be done in two steps. To con-
vert from RGB to HSI we first rotate the coordinates so that the third axis is along G =
R, B = Rline. This gives a rectangular coordinate version of the HSI space. The sec-
ond step is a change to cylindrical coordinates and proper scaling of the intensity. The
coordinates of the rectangular HSI space will be labeled M,, M>, I, where I, is the in-
tensity and M,, M, describe the hue-saturation plane. (See Figure 10-37.)
The transformation is

Diideuu- Or. «alps

[M;M,1;,]=[RGB]|-1AN6 1/N2 1/V3 (10.39)
-1N6 -1/N2 1/3
384 CHAPTERTEN

FIGURE 10-36
HSI coordinates.

The conversion to cylindrical coordinates in HSI is

H = arctan(M, / M,)
S = (Mi + M3)! (10.40)
I =1, V3
Conversion from HSI to RGB takes the inverse steps

M, = S sinH
M, = ScosH (10.41)
Low

|
red

blue m2
green %& 4

ml

FIGURE 10-37
Correspondence between RGB and
HSI coordinates.
 LIGHT. COLOR. AND SHADING 385

FIGURE 10-38
Black The cone of possible colors.

21V6 -1/V6. -1/N6.

[RG B]=[M;M2I;]]} 0 1/V¥2 -1/V2 (10.42)
LAZ: “hile SRS

When dealing with inks on paper, we sometimes use a color space with cylindrical
coordinates where the angle is hue, brightness is along the axis, and the radius is how
much of the colored ink is on the paper. This gives a cone of colors. At the point is
black, along the axis is saturation 0 (grays), and along the surface of the cone is the
most saturated color. As you move up or down, the amount of black ink changes. Mov-
ing out or in adds or removes color from the portion of the paper not covered by black.
The radial saturation measure is usually normalized so that it is 1 on the cone’s surface.
A value of | then specifies the maximum saturation achievable by the inks for the
given brightness. (See Figure 10-38.)
Sometimes hue, saturation, and value (HSV) are used to specify colors. Value 1s
the intensity of the maximum of the red, blue, and green components of the color.
Value has the same “‘feel’’ as brightness (as value decreases, the shade looks darker)
but is easier to calculate. Also to make the calculation easier, the hue can be approxi-
mated by the distance along the edge of a hexagon. The relation between HSV and
RGB is given by the next two algorithms:

10.13 Algorithm RGB-TO-HSV(R, G, B. H, S, V) Converts from red, green, blue

values in the range 0 to | to hue, saturation, and value
Hue is expressed in degrees
Arguments R, G, B the red, green, blue coordinates
H, S, V for return of the hue, saturation, and value
Local R1, Gl, B1, how relatively close the color is to red, green, and blue
X intensity of the minimum primary color
BEGIN
H<0;
386 CHAPTERTEN

find dominant primary color

V <— MAX(R, MAXG, B));
find the amount of white
X <— MIN(R, MIN(G, B)); .
determine the normalized saturation
S<(V —- X)/V;
IF S = 0 THEN RETURN;
R1:=(V — R)/(V - X);
G1 :=(V — G)/(WV — X);
BIs= (= By (WVa
in which section of the hexagon does the color lie?
IFR = VTHEN
IF.'G = XTHENH —5 + Bi
BESEA eG
ELSE IF G = VTHEN
IFB = XTHENH< RI ¢+ 1
ELSE H 3 Bil
BESE
IFR = X THEN H —3 Gl
BES Ee Siw Rl
convert to degrees
H <—H * 60
RETURN;
END;

The algorithm first finds the value parameter, which is simply the intensity of the
dominant primary color. Next, it finds the minimum primary color. This can be mixed
with equal amounts of the other two colors to form white and is therefore a measure of
how much white is in the color. The expression V — X is the amount of the dominant
color left after part is used to make the white component of the color. The relative
strength of this net color becomes our normalized saturation measure. If the saturation
is 0 then there is no hue (we have a gray), so the algorithm returns with a default hue of
0. The R1, Gl, and B1 variables are used to tell how much of the second most impor-
tant color there is relative to the dominant color. The IF statements then find the domi-
nant color and move the hue away from it by this amount.

10.14 Algorithm HSV-TO-RGB(H, S, V, R, G, B) Converts ae saturation, value

coordinates to red, green, blue ~

Arguments H, S, V the hue, saturation, and value

R, G, B for return of the red, green, blue coordinates
Local H1 the hue in units of 1 to 6
I integer part of the H1 hue, indicates the dominant color
F fractional part of the H1 hue, used to determine second color
A an array that holds the three color values while deciding which is which
BEGIN
convert from degrees to hexagon section
Hl —H/ 60;
find the dominant color
 LIGHT, COLOR, AND SHADING 387

I — INT[H1);
<a Hie ols
A{1] —V;
A[2] —V;
AD3] <= Vee dor (S« F));
A[4] —Vx(1 — S);
A[5] — A[4];
AO] < We Sel ="F)));
map strengths to rgb
IFI > 4THENI
<I — 4ELSEI<I + 2;
R—AI[I);
IF 1 > 4THENI<I — 4ELSEI <I + 2;
B < A{]]; i
IFI > 4THEN I <I — 4ELSEI <I + 2;
G <—Al]];
RETURN;
END;

In this algorithm the integer part of the hue determines the dominant color. Its in-
tensity V is saved in the first and second array cells. The fractional part describes the
second most important color. Its value is in either the third or sixth array cell, depend-
ing on which side of the dominant color it lies. The weakest color intensity is saved in
the fourth and fifth cells. Then the values are copied from the array to the R, G, B pa-
rameters according to which color was dominant (the original I).
Because of the simple way that HSV calculates hue, the colors are mapped onto
a hexagonal pyramid or hexcone instead of a circular cone. (See Figure 10-39.) A vari-
ation on the HSV model called the hue, lightness, and saturation model (HLS) is used
by Tektronix. It deforms the hexcode by stretching the white point along the value
axis. This forms two hexcones placed base to base, with black at one apex and white at

green yellow

cyan, red

magenta

FIGURE 10-39
black The HSV model hexcone of color.
388 CHAPTERTEN

the other. In this model, black has a lightness of 0 and white a lightness of 1, but fully
saturated colors have a lightness value of only 0.5. The lightness is defined as the aver-
age of the maximum and minimum RGB values, which is (V + X)/2 in the notation of
algorithm 10.13. The saturation is (V — X)/(V + X) if the lightness is less than 0.5,
and (V — X)/(2 — V — X) for lightness greater than 0.5. (See Figure 10-40.) The
HSI, HSV, and HLS color models are useful in implementing programs where the user
must select colors. It is not always easy to tell which RGB values are needed to get a
desired color. The HSV model corresponds to how an artist mixes colors: selecting a
hue, adding white to get the proper tint, and adding black to get the desired shade. It is
more intuitive. We might allow the user to specify a color using the HSV model and
then convert it to RGB for display. Or the user might select a color from a displayed
palette which would be converted to HSV units for later modification.

GAMMA CORRECTION
The above transformations are based on the fact that light behaves linearly, that the ef-
fect of light from two sources is simply the sum of the individual sources. While this is
true for light, it probably won’t be true for the voltages applied to a monitor to produce

white

*~ yellow

cyan red

magenta

FIGURE 10-40
black The HLS model double hexcone of color.
 LIGHT. COLOR. AND SHADING 389

the light. Doubling the voltage at the monitor will probably not yield twice the light.
Instead, there is typically a power relation.
l=kv’ (10.43)
The exponent + is usually around 2.
Suppose we determine the value (I) for an intensity we would like to display. We
still must determine the proper voltage setting for the display hardware to give this in-
tensity. To do this, we will have to solve Equation 10.43 for (V).

i l/y
V= (— (10.44)
k

This is called gamma correction. One fast way of performing gamma correction is to
construct a table of the supported intensity values and corresponding display voltages.
Gamma correction then becomes a simple table lookup. For a color display, gamma
correction must be applied to each of the three primary colors. Some monitors provide
internal circuitry to perform the gamma correction and thus appear to be linear de-
vices.

COLOR TABLES
Some display devices are capable of generating a large variety of colors. For example,
if a device has 8 bits of control over each of the red, green, and blue channels, then
each channel can have 256 intensity levels, and together 16,777,216 different colors
may be produced. To fully specify all these colors we would need 24 bits per pixel in
the frame buffer memory. Since we can usually get by with far fewer colors, this is not
very cost-effective. An alternative is a color table. A color table allows us to map be-
tween a color index in the frame buffer and a color specification. Suppose our frame
buffer had 8 bits per pixel. This would allow us only 256 colors; but what if we use the
frame buffer entry as an index to access a color table. Suppose the color table has 256
entries, each entry containing a 24-bit color selection. Then we can show any 256 of
the 16,777,216 possible colors. And by changing the values in the color table, we can
change the available color selection to contain the 256 colors we most desire for the
current scene. (See Figure 10-41.)
Note that changing a value in the color table can alter the appearance of large
portions of the display (any part of the display which referenced that entry). Because it
is the hardware which makes these changes, they usually occur very fast. Some clever
techniques have been devised which take advantage of this hardware feature. For
example, we might give the screen a background color from color-table entry 1 and
draw a complex object using color-table entry 2. Now if table entry 2 has the same
value as table entry 1, then the object will be invisible; it will be indistinguishable from
the background. But if we change table entry 2 to some other color, the object will sud-
denly appear. Changing its color can make it visible or invisible, and this change is
often far faster than we can draw or erase the object. Drawing or erasing requires
changing alot of pixel values in the frame buffer, while changing the color only re-
quires changing a single color entry.
390 cCHAPTERTEN

red green blue

wo
©
ne

8-bit index

255
FIGURE 10-41
256 X 24-bit color table A color table.

We can have several invisible objects (as many as there are color-table entries).
By sequencing through visibility changes, animation effects can be achieved.
Color tables have been used to give false color or pseudocolor to gray-scale im-
ages. They have also been used for gamma correction, lighting and shading, and color
model transformations.

EXTENDING THE SHADING MODEL

TO COLOR
As we mentioned at the start of the chapter, objects have a color-dependent index of re-
flection which can be represented by three reflectivity values: one for red, one for
green, and a third for blue. Our simple model for specular reflection is not color-depen-
dent, but the light sources can be colored. Background light and point sources must be
given three components. As a result, our shading expression of Equation 10.16 splits
into three equations, one for each color component.
PAIRKE NN) DG Fee
Ves BRet > Polk Ny PDS HENT
J Cir U; =

P,.[R,(L:N) + DGF/(E - N)]

Ng = BR,
ean + > = C+U, (10.45)

Vo = BeR, + > CU

j
 LIGHT, COLOR, ANDSHADING 39]

FURTHER READING
An introduction to the problems of realistic image generation is given in [NEW77].
First attempts at shading just used the distance from the viewpoint to determine bright-
ness [WYL67]. An early model for shading is described in [PHO75], and an alterna-
tive computation method for it was given in [DUF79]. This model was improved in
[BLI77]; it is the one used in this text. The model was further extended to include col-
ored specular reflections in [COO82]. Extensions for light sources with intensity distri-
butions and color are given in [WAR83]. Hardware extensions to the frame buffer have
been devised to allow fast scan conversion, hidden-surface removal, clipping, smooth
shading, shadows, and transparency. The method, which is based on all pixels being
able to evaluate linear expression in parallel, is described in [FUC85]. The idea of im-
aging transparent objects by modifying the painter’s algorithm was presented in
[NEW72]. Improved transparency approximations are discussed in [KAY79]. Gouraud
presented smooth shading by interpolation in [GOU71]. A comparison of shadow al-
gorithms is given in [CRO77]. Shadows are also discussed in [NIS85b]. Hidden-sur-
face techniques are used in [ATH78] to find polygons that are not in shadow. In
[WIL78] shadows are determined by first computing a Z buffer using the light source
as the viewpoint. Then as each point is calculated for display, it is compared against
that buffer to find out if anything lies in front relative to the light source to cause a
shadow. This allows curved shadows on curved objects. An extension of scan-line al-
gorithms to include shadows is given in [BOU70]. The method of generating shadows
by an extended Z buffer is described in [BRO84]. Ray tracing was first suggested in
[APP68]. Modeling of light and ray tracing are discussed in [WHI80]. An introduction
to ray tracing and its use with constructive solid geometry are given in [ROT82]. A sys-
tem for modeling, ray tracing, and display of three-dimensional objects is described in
[HAL83]; the system uses an improved illumination model including transparency.
Ray tracing for shadows and reflection of diffuse radiation are described in [NIS85a].
Ray-tracing research includes methods for antialiasing and also for the elimination of
sharp edges for shadows, motion blur, and translucency. See [AMA84], [COO84],
[HEC84], and [LEE85]. There is also research into ray tracing of objects other than
polygons. See [BLI82], [BOU85], [KAJ83a], [KAJ83b], [TOT85], and [VAN84].
And there is interest in improving the speed of ray tracing. See [AMA84], [GLA84],
[HEC84], [PLU85], and [WEG84]. Ray tracing can reproduce the specular reflection
of objects by objects, but does not model the interaction of diffuse radiation between
objects. This is considered in [GOR84] and [COH85]. Our system allows us to shade
surfaces composed of polygons of uniform color, but real objects can have textures and
variations of surface color. One approach to this problem is to map texture patterns
onto the surface. See [BLI76] and [CAT80]. Another is to define a three-dimensional
texture pattern which can be accessed for surface points [PEA85]. Texture patterns can
be used to provide perturbations to the surface normal as well as the reflectivity. This
can give surfaces a rough or wrinkled appearance. See [BLI78] and [HAR84]. Re-
views of techniques to obtain continuous tone images from bilevel devices are given in
[ALG83] and [KNO72]. Halftoning with an ordered dither is described in [JAR76],
[JUD74], [KLE70], and [LIP71]. Sometimes the halftone pattern is oriented at an
392 CHAPTERTEN

angle to the scan direction; [HOL79] describes how to determine an equivalent pattern
aligned with the scan direction. A more sophisticated halftone method than the one we
have presented includes the error, due to threshold test of a pixel value, in calculation
of the neighboring pixel’s value [FLO76]. A classic introduction to the ideas of hue,
value, and chroma is found in [MUNO5]. Color models are presented in [JOB78],
[LEH84], and [SMI78]. Some color models which we did not present strive to
parameterize the color space such that equal changes in parameters are perceived as
equivalent changes in color. See [MEY80] and [TAJ83]. The theory of exact color re-
production for light sources as in display monitors, transmission/absorption as in
photographic transparencies, and reflection absorption as in lithography is discussed in
[HOR84]. There are several reports on how to effectively use color tables. One ap-
proach is to choose colors to best cover the volume of the color space actually used in
the scene. See [HEC82] and [LEH84]. A method for determining table values which
compensate for nonlinearities in the hardware is presented in [CAT79]. By storing sur-
face normals as color-table indices, the lighting of a scene can be quickly changed by
altering the color table [BAS81]. Color-table techniques including animation are de-
scribed in [SHO79] and [SLO79]. Some rules for the effective use of color based on
the properties of vision are given in [MUR84].

[ALG83] Algie, S. H., ‘‘Resolution and Tonal Continuity in Bilevel Printed Picture Quality,’’ Computer Vi-
sion, Graphics, and Image Processing, vol. 24, no. 3, pp. 329-346 (1983).
[AMA84] Amanatides, J., ‘‘Ray Tracing with Cones,’’ Computer Graphics, vol. 18, no. 3, pp. 129-135
(1984).
[APP68] Appel, A., ‘‘ Some Techniques for Shading Machine Renderings of Solids,’’ AFIPS SJCC, vol. 32,
pp. 37-45 (1968).
[ATH78] Atherton, P., Weiler, K., Greenberg, D., ‘‘Polygon Shadow Generation,”’ Computer Graphics, vol.
12, no. 3, pp. 275-281 (1978).
[BAS81] Bass, D. H., ‘‘Using the Video Lookup Table for Reflectivity Calculations: Specific Techniques
ne oe Results,’’ Computer Graphics and Image Processing, vol. 17, no. 3, pp. 249-261
a :
{BLI76] Blinn, J. F, Newell, M. E., ‘‘Texture and Reflection in Computer Generated Images,’’
Communi-
cations of the ACM, vol. 19, no. 10, pp. 542-547 (1976).
[BLI77] Blinn, J. F., ‘‘Models of Light Reflection for Computer Synthesized Pictures,’’ Computer
Graphics,
vol. 11, no. 2, pp. 193-198 (1977).
{BLI78] Blinn, J. F, ‘‘Simulation of Wrinkled Surfaces,”’ Computer Graphics, vol.
12, no. 3, pp. 286-292
(1978).
[BLI82] Blinn, J. F, ‘‘Light Reflection Functions for Simulation of Clouds and Dusty Surfaces,’’
Computer
Graphics, vol. 16, no. 3, pp. 21-29 (1982). eee
[BOU70] Bouknight, J., Kelley, K., ‘An Algorithm for Producing Half-Tone Computer
Graphics Presenta-
tions with Shadows and Movable Light Sources,’’ AFIPS SJCC, vol. 36, pp. 1-10 (1970).
[BOU85] Bouville, C., ‘‘Bounding Ellipsoids for Ray-Fractal Intersection,”’ Computer
Graphics, vol. 19,
no. 3, pp. 45-52 (1985).
[BRO84] Brotman, L. S., Badler, N. I., “‘Generating Soft Shadows with a
Depth Buffer Algorithm,’’ JEEE
Computer Graphics and Applications, vol. 4, no. 10, pp. 5-12 (1984).
wean rok E., “A Tutorial on Compensation Tables,’’ Computer
Graphics, vol. 13, no. 2, pp. 1-7

[CAT80] Catmull, E., Smith, A. R., ‘‘3-D Transformations of Images

in Scanline Order,”’ Computer
Graphics, vol. 14, no. 3, pp. 279-285 (1980).
[COH85] Cohen, M. F, Greenberg, D. P., ‘‘The Hemi-Cube: A
Radiosity Solution for Complex Environ-
ments,’’ Computer Graphics, vol. 19, no. 3, pp. 31-40 (1985).
 LIGHT, COLOR, AND SHADING 393

[COO82] Cook, R. L., Torrance, K. E., ‘‘A Reflectance Model for Computer Graphics,’’ ACM Transactions
on Graphics, vol 1, no. 1, pp. 7-24 (1982).
[C0084] Cook, R. L., Porter, T., Carpenter, L., ‘‘Distributed Ray Tracing,’’ Computer Graphics, vol. 16,
no. 3, pp. 137-143 (1984).
[CRO77] Crow, F. C., ‘‘Shadow Algorithms for Computer Graphics,’’ Computer Graphics, vol. 11, no. 2,
pp. 242-247 (1977).
[DUF79] Duff, T., ‘‘Smoothly Shaded Renderings of Polyhedral Objects on Raster Displays,’’ Computer
Graphics, vol. 13, no. 2, pp. 270-275 (1979).
[FLO76] Floyd, R. W., Steinberg, L., ‘‘An Adaptive Algorithm for Spatial Grey Scale,’’ Proceedings of the
Society for Information Display, vol. 17, pp. 75-77 (1976).
[FUC85] Fuchs, H., et al., ‘‘Fast Spheres, Shadows, Textures, Transparencies, and Image Enhancements in
Pixel-Planes,’’ Computer Graphics, vol. 19, no. 3, pp. 111-120 (1985).
[GLA84] Glassner, A. S., ‘“Space Subdivision for Fast Ray Tracing,’’ IEEE Computer Graphics and Applica-
tions, vol. 4, no. 10, pp. 15-22 (1984).
[GOR84] Goral, C. M., Torrance, K. E., Greenberg, D. P., Battaile, B., ‘‘Modeling the Interaction of Light
between Diffuse Surfaces,’’ Computer Graphics, vol. 18, no. 3, pp. 213-222 (1984).
[GOU71] Gouraud, H., “‘Continuous Shading of Curved Surfaces,’’ IEEE Transactions on Computing, vol.
C-20, no. 6, pp. 623-629 (1971).
[HAL83] Hall, R. A., Greenberg, D. P., ‘‘A Testbed for Realistic Image Synthesis,’’ JEEE Computer
Graphics and Applications, vol. 3, no. 11, pp. 10-20 (1983).
[HAR84] Haruyama S., Barsky, B. A., ‘‘Using Stochastic Modeling for Texture Generations,’’ JEEE Com-
puter Graphics and Applications, vol. 4, no. 3, pp. 7-19 (1984).
[HEC82] Heckbert, P. S., ‘‘Color Image Quantization for Frame Buffer Display,’’ Computer Graphics, vol.
16, no. 3, pp. 297-307 (1982).
[HEC84] Heckbert, P. S., Hanrahan, P., ‘‘Beam Tracing Polygonal Objects,’’ Computer Graphics, vol. 18,
no. 3, pp. 119-127 (1984).
[HOL79]- Holladay, T. M., ‘‘An Optimum Algorithm for Halftone Generation for Displays and Hard
Copies,’’ SID Digest, vol. 10, pp. 102-103 (1979).
[HOR84] Horn, B. K. P, ‘‘Exact Reproduction of Colored Images,’’ Computer Vision, Graphics, and Image
Processing, vol 26, no. 2, pp. 135-167 (1984).
[JAR76] Jarvis, J. F, Judice, C. N., Ninke, W. H., ‘‘A Survey of Techniques for the Display of Continuous
Tone Pictures on Bilevel Displays,’’ Computer Graphics and Image Processing, vol. 5, no. 1, pp. 13-
40 (1976).
[JOB78] Joblove, G. H., Greenberg, D., ‘‘Color Spaces for Computer Graphics,’’ Computer Graphics, vol.
12, no. 3, pp. 20-25 (1978).
[JUD74] Judice, C. N., Jarvis, J. F, Ninke, W. H., ‘‘Using Ordered Dither to Display Continuous Tone Pic-
tures on an AC Plasma Panel,’’ Proceedings of the SID, vol. 15, no. 4, pp. 220-228 (1974).
[KAJ83a] Kajiya, J. T., ‘‘New Techniques for Ray Tracing Procedurally Defined Objects,’’ Computer
‘Graphics, vol. 17, no. 3, pp. 91-102 (1983).
[KAJ83b] Kajiya, J. T., ‘‘New Techniques for Ray Tracing Procedurally Defined Objects,’’ ACM Transac-
tions on Graphics, vol. 2, no. 3, pp. 161-181 (1983).
[KAY79] Kay, D. S., Greenberg, D., ‘‘Transparency for Computer Synthesized Images,’’ Computer
Graphics, vol. 13, no. 2, pp. 158-164 (1979).
[KLE70] Klensch, R. J., Meyerhofer, D., Walsh, J. J., ‘*Blectronically Generated Halftone Pictures,’’ RCA
Review, pp. 517-533 (Sept. 1970).
{KNO72] Knowlton, K., Harmon, L., ‘‘Computer-Produced Grey Scales,’’ Computer Graphics and Image
Processing, vol. 1, no. 1, pp. 1-20 (1972).
[LEE85] Lee, M. E., Redner, R. A., Uselton, S. P., ‘‘Statistically Optimized Sampling for Distributed Ray
Tracing,’’ Computer Graphics, vol. 19, no. 3, pp. 61-67 (1985).
[LEH84] Lehar, A. F, Stevens, R. J., ‘‘High-Speed Manipulation of the Color Chromaticity of Digital Im-
ages,’’ IEEE Computer Graphics and Applications, vol. 4, no. 2, pp. 34-39 (1984).
[LIP71] Lippel, B., Kurland, M., ‘‘The Effect of Dither on Luminance Quantization of Pictures,’ JEEE
Transactions on Communication Technology, vol. COM-19, no. 6, pp. 879-888 (1971).
[MEY80] Meyer, G. W., Greenberg, D. P,, ‘*Perceptual Color Spaces for Computer Graphics ,’’ Computer
Graphics, vol. 14, no. 3, pp. 254-261 (1980).
394 CHAPTERTEN

[MUNOS5] Munsell, A. H., A Color Notation, 12th ed., Munsell Color Company, Inc., Baltimore, Md.
(1971).
[MUR84] Murch, G. M., ‘‘ Physiological Principles for the Effective Use of Color,’’ IEEE Computer
Graphics and Applications, vol. 4, no. 11, pp. 49-54 (1984). .
[NEW72] Newell, M. E., Newell, R. G., and Sancha, T. L., ‘‘A Solution to the Hidden Surface Problem,”’
Proceeding of the ACM Annual Conference, pp. 443-450 (1972).
[NEW7Z7] Newell, M. E., Blinn, J. F, ‘“The Progression of Realism in Computer Generated Images,’’ ACM
“ 77, Proceedings of the Annual Conference, pp. 449-451 (1977).
[NIS85a] Nishita, T., Nakamae, E., ‘‘Continuous Tone Representation of Three-Dimensional Objects Taking
Account of Shadows and Interreflection,’’ Computer Graphics, vol. 19, no. 3, pp. 23-30 (1985).
(NIS85b] Nishita, T., Okamura, I., Nakamae, E., ‘‘Shading Models for Point and Linear Sources,’’ ACM
Transactions on Graphics, vol. 4, no. 2, pp. 124-146 (1985).
[PEA85] Peachey, D. R., “‘Solid Texturing of Complex Surfaces,’’ Computer Graphics, vol. 19, no. 3, pp.
279-286 (1985).
[PLU85] Plunkett, D. J., Bailey, M. J., ‘“The Vectorization of a Ray-Tracing Algorithm for Improved Execu-
tion Speed,’’ JEEE Computer Graphics and Applications, vol. 5, no. 7, pp. 52-60 (1985).
[PHO75] Phong, Bui-Tuong, ‘‘Illumination for Computer-Generated Pictures,’’ Communications of the ACM,
vol. 18, no. 6, pp. 311-317 (June 1975).
[ROT82] Roth, S. D., ‘‘Ray Casting for Modeling Solids,’’ Computer Graphics and Image Processing, vol.
18, no. 2, pp. 109-144 (1982).
[SHO79] Shoup, R. G., “‘Color Table Animation,’ Computer Graphics, vol. 13, no. 2, pp. 8-13 (1979).
[SLO79] Sloan, K. R., Brown, C. M., *‘Color Map Techniques,’’ Computer Graphics and Image Process-
ing, vol.10, no. 4, pp. 297-317 (1979).
[SMI78] Smith, A. R., ‘‘Color Gamut Transform Pairs,’’ Computer Graphics, vol. 12, no. 3, pp. 12-19
(1978).
[TAJ83] Tajima, J., ‘‘Uniform Color Scale Applications to Computer Graphics,’’ Computer Vision, Graphics,
and Image Processing, vol. 21, no. 3, pp. 305-325 (1983).
[TOT85] Toth, D. L., ‘‘On Ray Tracing Parametric Surfaces,’ Computer Graphics, vol. 19, no. 3, pp. 171-
179 (1985).
[VAN84] Van Wijk, J. J., ‘Ray Tracing Objects Defined by Sweeping Planar Cubic Splines,’’ ACM Transac-
tions on Graphics, vol. 3, no. 3, pp. 223-237 (1984).
[WAR83] Warn, D. R., ‘‘Lighting Controls for Synthetic Images,’’ Computer Graphics, vol. 17, no. 3, pp.
13-21 (1983).
[WEG84] Weghorst, H., Hooper, G., Greenberg, D. P., ‘‘Improved Computational Methods for Ray Trac-
ing,’’ ACM Transactions on Graphics, vol. 3, ne. 1, pp. 52-69 (1984).
[WHI80] Whitted, T., ‘‘An Improved Illumination Model for Shaded Display,’’ Communications of the ACM,
vol. 23, no. 6, pp. 343-349 (1980).
[WIL78] Williams, L., ‘‘Casting Curved Shadows on Curved Surfaces,”’ Computer Graphics, vol. 12, no.
3, pp. 270— 274 (1978).
[WYL67] Wylie, C., Romney, G., Evans, D., Erdahl, A., ‘‘Half-Tone Perspective Drawings by Computer,”’
AFIPS FJCC, vol. 31, pp. 49-58 (1967).
i

EXERCISES
10-1 Suppose that a rectangular polygon is drawn with vertices at (0:0), =) Ody, = Gas
0), (1, 0, 0), and is viewed with the default parallel projection. Give the shading value
which
would result from each of the following parameter settings.
a) Default
b) SET-OBJECT-SHADE(0.4, 10, 10)
SET-LIGHT(0.5, 0.5, 1, 1, 0)
c) SET-OBJECT-SHADE(0.4, 10, 10)
SET-LIGHT(0.5, 0.5, 1, 1, 1)
 LIGHT, COLOR. AND SHADING 395

d) SET-OBJECT-SHADE(0.4, 2, 10)
SET-LIGHT(0.5, 0.5, 1, 1, 0)
e) SET-OBJECT-SHADE(0.4, 2, 10)
SET-LIGHT(— 1, 0.5, —0.5, 1,0)
10-2 State the meaning of each of the symbols found in Equation 10.16.
10-3 Extend the shading model to include the display of objects which emit their own light as
well as illuminated objects.
10-4 Could the simple Z-buffer hidden-surface algorithm be used with transparent objects?
Why?
10-5 How many intensity levels can be formed from an n X npixel cell if each individual pixel
can have m intensity settings?
a) Assuming the pixel intensities are uniformly spaced
b) Assuming the pixel intensities are arbitrary (e.g., an intensity lookup table is used)
10-6 Determine the RGB to XYZ transformations matrix for the NTSC standard primaries using
the definition of white given in Equation 10.29.
10-7 Develop algorithms to convert between the RGB color model and the HLS color model.
10-8 Identify the x,y value for three colors which will maximize the area of the color gamut
triangle on the CIE diagram. What is the area of this triangle? Estimate the area covered by all
visible colors on the diagram and determine the area of colors which cannot be produced with
only three primaries. Compare the area of your maximum triangle to that for the gamut of a
color monitor.
10-9 For some luminancy, describe the surface in RGB space which corresponds to the color
gamut of the normalized x,y values.

PROGRAMMING PROBLEMS
10-1 Implement algorithms 10.1 through 10.11 to add automatic shading to your graphics
system.
10-2 Test your shading routines by constructing a house, selecting a view which shows at least
three surfaces, and generating several images, each of which differs from the others by a change
of a shading parameter. Observe the effect of each parameter independently in this manner.
10-3 Write the routines
INQUIRE-SHADING(ON-OFF)
INQUIRE-LIGHT(X, Y, Z, BRIGHTNESS, BACKGROUND)
INQUIRE-OBJECT-SHADE(REFLECTIVITY, REFRACTION, GLOSS)
which return to the user the current shading parameter settings.
10-4 Extend the program of Programming Problem 9-8, which gives successive views of an ob-
ject to show the object shaded.
10-5 Write a program which gives successive images of a shaded object drawn by procedure
FOO within the cube 0 < X < 1,0 <Y< 1,0 <Z< 1. The default viewing parameters wiil al-
ways be used, but in successive pictures, the light source will appear to rotate around the object.
10-6 Extend the interactive display program of Programming Problem 8-14 to include setting of
the light-source parameters.
10-7 Extend the graphics system to handle transparent objects. Do this by modifying the SET-
OBJECT-SHADE algorithm to include a transparency coefficient, and define a display-file style
command for transparency. Finally, modify the FILLIN algorithm for polygons so that when
396 cCHAPTERTEN

filling a transparent polygon, each pixel value is set to the current shade plus the transparency
times the old shade value of the pixel.
10-8 Extend the graphics system to show shaded images via halftoning.
*10-9 Extend the graphics system to include the Gouraud smooth shading interpolation for
shades specified by the user. Polygon-entry routines should be extended to include an intensity
value for each vertex. Clipping routines must supply intensity values for new vertices. The
display file must be able to save these intensity values. And finally, the polygon-filling routine
must interpolate over the shades.
 CHAPTER

ELEVEN
CURVES
AND
FRACTALS

INTRODUCTION
Throughout this book we have dealt exclusively with straight-line segments. We saw in
the first chapter that straight-line segments are fairly easy to generate. In later chap-
ters, we learned how to represent and manipulate line segments. Straight lines have a
simple mathematical form, which makes them easy to deal with in such operations as
transformations or clipping. The real world, however, is not made of stick figures.
Many of the things which we might like to model and display involve curves. We
might like to plot a mathematical function or the path of a rocket, or we might wish to
design the hood of a sports car or the wing of an airplane, or we might desire to draw a
woman’s face. Natural objects are neither perfectly flat nor smoothly curved but often
have rough, jagged contours. In this chapter we shall introduce methods for generating
curved lines. The discussion will be rather superficial, since a thorough study is
beyond the scope of this book. Nevertheless, we shall develop spline interpolation al-
gorithms for drawing curved lines and fractal curve generators for rough, natural-look-
ing objects.

CURVE GENERATION
There are two approaches which can be taken to draw curved lines. One approach is to
develop a curve generation algorithm such as a DDA. With this approach, a true curve is
created. The second method is to approximate a curve by a number of small straight-

397
 398 CHAPTER ELEVEN

line segments. The interpolation techniques which we shall discuss later are used for
this second approach. We shall begin our discussion by considering the true-curve gen-
eration method.
Suppose that we have a curve, such as the arc of a cirole, that we wish to draw. If
we know a differential equation for this curve, we can write a digital differential
analyzer algorithm which will calculate the coordinates of points on the curve. We can
then change the intensity values of the pixels which contain these points, and an image
of the curve will appear on the screen.
The differential equations for simple curves such as circles and ellipses are fairly
easy to solve, and generation algorithms for them can be implemented in hardware. A
display device may be capable of drawing arcs as well as line segments.
Let us step through the development of a DDA algorithm for generating circular
arcs. The equations for the arc coordinates can be written in terms of an angle parame-
ter A as follows:

xX = RcosA + Xg
Cid)
y =RsinA + yo

where (Xo, Yo) is the center of curvature, and R is the arc radius. (See Figure 11-1.)
Differentiating tells us to change x by dx and y by dy if we change A by dA.

dx = — RsinAdA
(11.2)
dy = RcosAdA

We notice that Equation 11.1 gives

RcosA = x — Xo
(11.3)
RsinA
= y — yo
so that

FIGURE 11-1
(Xo, Yo) Parameters for an arc.
 CURVES AND FRACTALS 399

dx = — (y — yo)dA
(11.4)
dy = (x — Xo)dA
Now the changes dx and dy tell us the amount to add to an old point on the arc to get a
new point.

X= y+ OX = x, —(y, —"yojdA
(11.5)
You=ey wathidyase Yui aon iX)dA
These equations form the basis for an arc generation algorithm. In this algorithm, we
start with the center of curvature (Xo, Yo), the total angle of the arc to be drawn A, a
point at which the arc drawing should begin (x, y), and the intensity at which to set the
pixels. The step size of the parameter dA should be small enough not to leave gaps in
the arc and small enough to give a good approximation to a circle. The arc-generating
algorithm is as follows:

11.1 Algorithm ARCDDA(X0, YO, A, X, Y, INTENSITY) Counterclockwise circular

arc generation
Arguments XO, YO the center of curvature
A the arc angle
X, Y the starting point for the arc drawing
Global FRAME the frame buffer array
Local XARC, YARC the next point to draw
A-DRAWN the amount of angle currently covered
DA the angle increment
Constant ROUNDOFF some small number greater than any round-off error
BEGIN
IF| X — X0| + |Y — YO| < ROUNDOFF THEN RETURN;
A-DRAWN < 0;
find a suitable angle increment
DA < MIN(0.01, 1 / (3.2 « (IX — XO| + /Y — YO))));
set the first point of the arc
XARC < X;
WARCGKSN;
* generate the arc until desired angle is covered
WHILE A-DRAWN < A DO
BEGIN
find new point
XARC < XARC + (YO — YARC) * DA;
YARC < YARC + (XARC — X0) « DA;
A-DRAWN < A-DRAWN + DA;
set the corresponding pixel
FRAME[INT(XARC), INT(YARC)] < INTENSITY;
END;
RETURN;
END;

There are, however, some problems which arise when curves are generated at
this level. To specify a curve, we need more information than just its endpoints. This
400 = cHAPTER ELEVEN

may mean that a different display-file structure should be used. Another problem can
arise from transformations. A line segment when scaled is still a line segment, but
other curves may behave differently. For example, a circle when scaled in only one di-
rection becomes an ellipse. If we had only a circular-arc generator, we would be very
limited in our ability to scale pictures. Still another problem is the fact that our clip-
ping algorithm works only for points and straight lines. If we wished to clip arcs to
some window boundary, a new clipping algorithm would have to be developed. Al-
though these problems are not insurmountable, there is yet another limitation to the
curve generation approach; not every curve which we wish to draw has a simple gener-
ation algorithm. Curves for airplane wings or sports car hoods or human faces are not
simple circles or ellipses. We would have to approximate such curves by patching to-
gether pieces of circles or ellipses or spirals. But we might just as well patch together
many small straight-line segments and then be able to use the display file and the trans-
formations and the clipping algorithm which we have studied.

INTERPOLATION
How can we express a curve which has no simple mathematical definition? We can
draw an approximation to such a curve if we have an array of sample points. We can
then guess what the curve should look like between the sample points. If the curve is
smooth and our samples are close enough together, we can make a pretty good guess as
to what the missing portions should look like. Our guess will probably not be exactly
right, but it will be close enough for appearances. We fill in portions of the unknown
curve with pieces of known curves which pass through the nearby sample points. Since
the known and unknown curves share these sample points in a local region, we assume
that in this region, the two curves look pretty much alike. We fit a portion of the un-
known curve with a curve that we know. Now we can fill in a gap between the sample
points by finding the coordinates of points along the known approximating curve and
connecting these points with line segments. (See Figure 11-2.)
The next thing which we might ask is what we should use for the known curve.
What is the mathematical expression for a function which passes through the sample
points in a limited region? There are many function forms which can be made to pass
through sample points by adjusting parameters. Polynomial, trigonometric, exponen-
tial, and other classes of functions have been used. Furthermore, there are many op-
tions within each class. We shall use polynomial functions to approximate the curve.
Functions are often expressed in the form y = f(x), but we shall prefer a
parametric form. >
x = fx(u)

y = fy(u) (11.6)
z = fz(u)
There are a couple of reasons why the parametric form is preferable. For one
thing, the difference between two and three dimensions is just the addition of a third
equation for z. The parametric form treats all three directions equally, and it allows
 CURVES AND FRACTALS 401

Unknown curve

°
° °

° o Known sample points

° °

° Fit a region with

a known curve
° °

° Calculate more points

from known curve
° °

cee ° Actually draw straight

. line segments connecting FIGURE 11-2
int ; ;
° ° Rae | The interpolation process.

multiple values (several y or z values for a given x) so that curves can double back or
even cross themselves. (See Figure 11-3.)
Now let’s invent a function which can be used for interpolation. There will be a
number of things wrong with this first attempt, but it will nevertheless serve to illus-
trate the approach.
Suppose we want a polynomial curve that will pass through n sample points.
(x), Yi Zi), (Xp, y2, Z>), see y Xs Yn: ze)

We shall construct the function as the sum of terms, one term for each sample point.
402 CHAPTER ELEVEN

FIGURE 11-3
The parametric form allows representation of curves which double back or cross themselves.

n
fx(u) =.) SeBia)
i=!

fy(u)= 2% y;Bi(u) (11.7)

ba

n
fz(u) = % z,B,(u)
1a

The functions B,(u) are called blending functions. For each value of u, they determine
how much the ith sample point affects the position of the curve. We can think of each
sample point as trying to pull the curve in its direction. The B;(u) function tells us how
hard the ith sample point is pulling. If for some value of u, B;(u) = 1 and for each j ¥
i, B,(u) = O, then the ith sample point has complete control of the curve. The curve
will pass through the ith sample point. If at a different value of u, one of the other sam-
ple points has complete control, then the curve will pass through this point as well.
What we want then are blending functions which for different values of u will give
control of the curve to each of the sample points in turn. The particular values of the
parameter u at which the curve passes through each of the sample points are not very
important, just so long as the sample points get control of the curve in proper order. We
shall create blending functions for which the first sample point (x,, y;, z,) has com-
plete control when u = —1, the second when u = 0, the third when u = 1, and so on.
The reason for this selection of u values is to make the actual implementation a bit sim-
pler. For this choice, we need for B,(u) a function which is lvatu = —1, and 0 for
u = 0, 1,2, ...,n — 2. An expression which is 0 at the correct places is

(ue 1) (Ue erate (Ne i)

Atu = —1Jitis

Cl) 2)C2S)3.CL. 2)
So dividing by this constant gives us 1 atu = —1, and this has just the behavior we
wish
 CURVES AND FRACTALS 403

ae u(u — 1)\u(Cu - 2)...fu-(—2

ys. [=n »)) (11.8)
kN
=2) =3 ell fi)
The ith blending function can be constructed in the same way to be 1 at u = i — 2,
and 0 at the other integers.

B,(u) wegUl _ tus)

en) .ee. [[us a
'Ge= ][u —
3))fu=— (G = u
1)... pa[w= @ = 2) ] (11.9)
Greed Sei eS) ee CLC
— 2b) 22 mn)
Using these expressions to approximate the curve is called Lagrange interpolation.
Let’s consider the case where there are four sample points. Then we will need
four blending functions. The above definition gives

sae u(u = Wine(Cu 2)

CH= 2S 3)

Bie) (u + iu1)(u — Lu)(u - 2 )

Cl) h=2)
(11.10)
w+ Duta = 2)
B es

oe (aN —1
Bite Cae Lye)

Penh CIN)
Using these functions and four sample points, we can construct a curve which passes
through the four sample points.

xX = x,B,(u) + xB,(u) + x3B3(u) + x4B,(u)

y = y,B,(u) + Y2B,(u) + y3B3(u) + y4B4(u) (11.11)

z = 2,B,(u) + z,B(u) + z3B3(u) + 24B,(u)

Now We expect this curve to lie fairly close to the actual curve in the region of these
four points, particularly in the middle, between the second and third sample points.
This is the portion of the curve followed as u varies between O and 1. We can find
The
points on the approximation curve by plugging in values for u (between 0 and 1).
results of these calculations provide coordinates which can be used as the endpoints for
approximati on to the approximati on
small line segments which we shall draw as an
the sec-
curve. For example, suppose that we wish to approximate the curve between
straight-line segments. We have as sample points
ond and third sample points by three
= 1. If we use our equations to calculate points atu = 1/3
the values at u = 0 and u
will have four points which will serve as the endpoints of the three
and at u = 2/3, we
this pro-
approximating segments. The entire curve can be approximated by repeating
the curve be-
cess. We take four consecutive sample points (1, 2, 3, 4) and approximate
picking up a new sam-
tween the middle two (2, 3). Then we step up one sample point,
404 CHAPTER ELEVEN

ae
itor

U=2 FIGURE 11-4

Calculating approximation points at
values of u between 0 and 1.

ple point at one end and discarding a point at the other end (2, 3, 4, 5). We can then ap-
proximate the next portion of the curve (3, 4). We continue moving through the sample
points until the curve is drawn. The initial and the final portions of the curve will re-
quire special handling. For the first four points (1, 2, 3, 4), we shall need to draw the
region between points 1 and 2 with u values between — | and 0. Likewise, the blend-
ing functions for the very last step of the curve should be evaluated with u values be-
tween | and 2. There is nothing particularly difficult about these special cases; we just
must handle them separately.

INTERPOLATING ALGORITHMS
Before discussing the shortcomings of these blending functions and presenting an im-
proved set, let us see how this curve-drawing program might be implemented. The first
thing which we notice is that the same blending function values are needed for each
section of the curve that is drawn. If each section is approximated by three straight-line
segments, then each section will require the blending function values for u at 0, 1/3,
2/3, and 1. (See Figure 11-4.)
These values can be calculated once and saved in an array for use in drawing
each curve section. The first algorithm which we shall present allows the user to
specify how many straight-line segments should be used to complete a section of the
curve. This number (NUMBER-OF-LINES) will tell us the u values at which to calcu-
late and save the blending function values. The value of B,(u) is stored in the BLEND
array at BLEND[I, U]. The blending function values needed for the first and last sec-
tions of the curve are also calculated and stored in arrays FIRST-BLEND and LAST-
BLEND. ~ .

11.2 Algorithm SET-SMOOTH(NUMBER-OF-LINES) User routine to set the

number of line segments per curve section
Argument NUMBER-OF-LINES the number of line segments per curve section
Global LINES-PER-SECTION storage for the NUMBER-OF-LINES specification
BLEND, FIRST-BLEND, LAST-BLEND
arrays size 4 by MAX-NUMBER-OF-LINES for blending function values
Local U, V, W parameters for blending function evaluation
1a variable for stepping through needed curve points
 CURVES AND FRACTALS 405

Constant MAX-NUMBER-OF-LINES the maximum allowed number of line seg-

ments per curve section
BEGIN
IF NUMBER-OF-LINES < 1
OR NUMBER-OF-LINES > MAX-NUMBER-OF-LINES THEN
RETURN ERROR ‘INVALID NUMBER OF LINE SEGMENTS’:
LINES-PER-SECTION <- NUMBER-OF-LINES;
FOR I = 1 TO NUMBER-OF-LINES DO
BEGIN
blending function values for the middle section
U <—I1/ NUMBER-OF-LINES;
BLEND[1, I] — U *(U — 1) *(U — 2)/(-6);
BLEND[2, I] —(U + 1) *(U -— 1) *(U — 2)/2:;
BLEND[3, I] —(U + 1) * U*(U — 2)/(—2);
BLEND(4, I] —(U + 1) *U*(U — 1)/6;
V.=—U.=-1;
blending function values for the first section of the curve
FIRST-BLEND[1, I] —V * (V — 1) * (V — 2)/(—6);
FIRST-BLEND[2, I] —(V + 1) *(V — 1) *(V — 2)/2;
FIRST-BLEND{3, I] —(V + 1) * V*(V — 2)/(-2);
FIRST-BLEND[4, I] —(V + 1) *V*(V — 1)/6;
W<U +1;
blending function values for the last section of the curve
LAST-BLEND[1, I] —W * (W — 1) * (W — 2)/(-6);
LAST-BLEND[2, I] — (W + 1) * (W — 1) * (W — 2)/2;
LAST-BLEND{3, I] <— (W + 1) * Wx* (W — 2)/(—2);
LAST-BLEND[4, I] — (W + 1) «Wx (W — 1)/6;
END;
RETURN;
END;

The next routine which we shall write multiplies the sample points and blending
functions to give points on the approximation curve. These points are then connected
by line segments. The sample points are assumed to be in the global arrays XSM,
YSM,-and ZSM, while the array of blending function values is passed as an argument.

11.3 Algorithm MAKE-CURVE(B) A routine to fill in a section of the curve

Argument B a4 by LINES-PER-SECTION array of blending function values
Global LINES-PER-SECTION the number of line segments per curve section
XSM, YSM, ZSM four-element arrays containing the sample points
Local I for stepping through the four sample points
J for stepping through the line segments of the curve section
X, Y, Z the coordinates of a point on the approximation curve
BEGIN
FOR J = 1 TO LINES-PER-SECTION DO
BEGIN
X — 0;
Y<0;
406 CHAPTER ELEVEN

Lic 0:
sum the contribution from each sample point
FOR I = 1TO4D0
BEGIN N
X<—X + XSM{[]]J * BUI, J);
Y<Y + YSM[I] * BE, J);
Z<—Z + ZSM[I] * BUI, J];
END;
draw the approximation line segment
LINE-ABS-3(X, Y, Z);
END;
RETURN;
END;

After a section of the curve has been drawn, we will shift the sample points so
that our blending functions can be applied to the next section. This is accomplished by
the following routine.

11.4 Algorithm NEXT-SECTION Routine to shift the sample points to get ready for
next curve section
Global XSM, YSM, ZSM the four-element sample point arrays
Local I a variable for stepping through the sample points
BEGIN
FOR I = 1TO3 DO
BEGIN
XSM[I] — XSM[I + 1];
YSM[I] — YSM[I + 1];
ZSM[I] — ZSM[I + 1);
END;
RETURN;
END;

To start drawing the curve, we require the first four sample points. With these
point values, we can use the FIRST-BLEND and BLEND arrays to draw the first two
curve sections. The START-CURVE algorithm does this. It expects as arguments ar-
rays containing the first four sample points. It loads these points into the XSM, YSM,
and ZSM arrays, the pen is positioned at the first sample point, and the MAKE-
CURVE routine is used to draw the first two sections of the curve. The sample points
are then shifted to prepare for drawing the next curve section:

11.5 Algorithm START-CURVE(XA, YA, ZA) User routine to start drawing a curve
Arguments XA, YA, ZA four-element arrays containing the first four sample points
Global XSM, YSM, ZSM four-element arrays for holding the sample points
FIRST-BLEND, BLEND
4 by LINES-PER-SECTION arrays containing blending function values
Local I for stepping through the sample points
BEGIN
load the sample points
FOR I = 1TO4DO
 CURVES AND FRACTALS 407

BEGIN
XSM[I] <— XA{IJ;
YSM[I] <— YA[I];
ZSM[I] — ZA[I];
END;
move to the start of the curve
MOVE-ABS-3(XA[1], YA[1], ZA[1]);
draw the first two sections
MAKE-CURVE(FIRST-BLEND);
MAKE-CURVE(BLEND);
prepare for the next section
NEXT-SECTION;
RETURN;
END;

Once the curve has been started, new sections can be added one at a time. (See
Figure 11-5.)

a
°
oa
7Os

7
°
°
/
7
ae

\
\ /
Nhs
\ /
vr

rh

/
/
p
/
/
if
FIGURE 11-5
2 Drawing a curve by interpolating over consecutive four-point
N regions.
408 cHAPTER ELEVEN

For each new sample point, a new section of the curve can be drawn; but since
we are adding the fourth sample point while interpolating between the second and third
sample points, the section of the curve being drawn always lags one sample point be-
hind the points entered. The following is a utility routine to place a new sample point
in the XSM, YSM, ZSM arrays.

11.6 Algorithm PUT-IN-SM(X, Y, Z) Routine to place a new sample point in the

XSM, YSM, ZSM arrays
Arguments X, Y, Z the new sample point
Global XSM, YSM, ZSM four-element arrays for saving sample points
BEGIN
XSM[4] < X;
YSM[4] —Y;
ZSM[4] <— Z;
RETURN;
END;

The routine to extend the curve takes a new sample point as its argument and
stores it in the XSM, YSM, ZSM arrays. It then uses the MAKE-CURVE routine to
draw the next curve section. Finally, it shifts the sample points to make ready for the
next curve section. The algorithm which does this is called CURVE-ABS-3.

11.7 Algorithm CURVE-ABS-3(X, Y, Z) User routine to extend the curve

Argument X, Y, Z the new sample point for the curve
Global BLEND 4 by LINES-PER-SECTION array of blending function values
BEGIN
PUT-IN-SM(X, Y, Z);
MAKE-CURVE(BLEND);
NEXT-SECTION;
RETURN;
END;

The curve may be extended as desired by repeated calls to CURVE-ABS-3, but

when we are ready to end it, we must process the last section specially. This is done by
the END-CURVE routine, which takes as an argument the last point on the curve. (See
Figure 11-6.) -

11.8 Algorithm END-CURVE(X, Y, Z) User routine to end*a curve

Argument X, Y, Z the last point on the curve.
Global BLEND, LAST-BLEND
4 by LINES-PER-SECTION arrays containing blending function values
BEGIN
PUT-IN-SM(X, Y, Z)
MAKE-CURVE(BLEND);
MAKE-CURVE(LAST-BLEND);
RETURN;
END;
 CURVES AND FRACTALS 409

FIGURE 11-6
Interpolation smoothing.

INTERPOLATING POLYGONS
The sides of a polygon can also be rounded by means of our blending functions. In
fact, a polygon is conceptually easier to deal with, since no special initial or final sec-
tions occur. We just step around the polygon, smoothing out each side by replacing it
with several small line segments. We start out with a polygon that has only a few sides
and end up with a polygon which has many more sides and appears smoother. (See Fig-
ure 11-7.)
The system will have some limit to the total number of sides a polygon can have.
The algorithm SMOOTH-POLY-ABS-3 given below smooths a polygon. It assumes
that a call has been made to the SET-SMOOTH routine to set up the proper BLEND
array values and that the system is capable of handling the resulting polygons. The re-
sulting polygon size will be the number of sides in the original polygon times the
LINES-PER-SECTION factor. The algorithm steps through the sides of the original
polygon and builds the new smoother polygon, which is drawn.

11.9 Algorithm SMOOTH-POLY-ABS-3(AX, AY, AZ, M) User routine to draw a

smoothed polygon
Arguments AX, AY, AZ arrays containing vertices of original polygon
- M the original number of polygon sides
Global BLEND the array of blending function values
LINES-PER-SECTION the number of line segments per original polygon
side
Local BX, BY, BZ arrays to hold smoothed polygon vertices
NSIDES number of sides for the smoothed polygon
J index variable for saving the smoothed polygon sides
K to step through the small segments for each polygon side
I, Il variables for stepping through the four sample points
L for stepping through the four blending functions
BEGIN
IF M < 3 THEN RETURN ERROR ‘POLYGON SIZE ERROR’;
NSIDES < LINES-PER-SECTION * M;
IF LINES-PER-SECTION = 1 THEN POLYGON-ABS-3(AX, AY, AZ, M)
Bis
410 CHAPTER ELEVEN

BEGIN
J<1;
FOR I1 = 1TOM DO
smooth all sides
FOR K = 1 TO LINES-PER-SECTION DO
smooth a side of the original polygon
BEGIN
BX[J] —0;
BY[J] —0;
BA <0:
I<Il;
FOR L = 1TO4DO
one side of the smooth polygon
BEGIN
BX(J] — BX[J] + AX[I] * BLEND[L, K];
BY[J] ct4 — BY[J] + AY[I] * BLEND[L, K];
BZ[J] — BZ[J) + AZ[I] * BLEND[L, K];
IF-1= MTHEN Ti V ELSE Seal;
END;
eet
END;
draw the result
POLYGON-ABS-3(BX, BY, BZ, NSIDES);
END;
RETURN;
END;

B SPLINES
Now that we have implemented a Lagrange interpolation program using our home-
made blending functions, let’s examine some of its inadequacies. One thing to notice
is that the sum of the blending functions is not | at every value of u. The blending
functions were designed to sum to | at integer values of u, but not at fractional values.
What does this mean? Suppose that all sample points had the same x value, x = X,.

FIGURE 11-7
Smoothing of a polygon.
 CURVES AND FRACTALS 411

We would then desire the approximating curve to also have constant values for x at the
points in between the sample points, but the approximating curve is

n
X mya x;B,(u) (tt t2)
1=

which for x; = x, gives

n
X=x, Bu) (11.13)
i=1
We will get the flat behavior we desire only if the sum of the blending functions is 1
for all values of u. This shortcoming also means that curves which should lie in a plane
may in fact wiggle in and out of the plane. This particular problem might be patched
by dividing the blending function values calculated in SET-SMOOTH by their sum for
each value of u, thereby normalizing the functions. But, alas, this is not the only prob-
lem. Each section of the curve is connected to the next section at a sample point, but
the slopes of the two sections need not match at this point. This means that there can
be corners at the sample points and that we may not have a completely smooth curve.
Finally, in pulling the curve all the way over to pass through one of the sample points,
we had to reduce the effect of all other points to zero; but as we leave the vicinity of
the sample point, all pointsin the region have some control overthe curve. Thus, we see
the control of the curve by a sample point pulsing in and out as we move along in u.
A more natural behavior might be to have a sample point’s control vary smoothly
from zero far away from the point to a maximum near the point. We can do this if we
don’t try to force the curve through the sample point, but rather gently pull it into the
neighborhood of the sample point. The result will be a curve which follows the general
contours indicated by the sample points but may not actually pass through the points
themselves. A set of blending functions which take this approach and also always sum
to 1 are called B splines. They generate curve sections which have continuous slopes so
that they fit together smoothly. (See Figure 11-8.)
The derivation of the B-spline functions is beyond the scope of this book, but we
shall present and implement the cubic B-spline functions, which are adequate for most

FIGURE 11-8
Smoothing using B splines.
412 CHAPTER ELEVEN

applications. The cubic B-spline blending functions interpolate over four sample points
and are cubic polynomials in u, just as our homemade blending functions. The cubic
B-spline functions for the middle of the curve are as follows:
B,(u) = (1 — u)7/6
B,(u) = uw?/2 — wu? + 2/3
(11.14)
B,(u) = — w/2 + u?/2 + u/2 + 1/6
B,(u) = u°/6
Starting or stopping the cubic B-spline blending functions requires two special curve
sections. The following functions should be used for the very first section of the curve.

By(u) = (1 — up
B3(u) = 21u°/12 — 9u*/2 + 3u
Bi(u) = —1lu?/12 + 3u2/2 go
B,(u) = u°/6
The last section of the curve would reverse these functions.

Big(u) = B41 — u)

Bor(u) = B3(1 — u)
(11.16)
B3p(u) = B31 — u)

Bar(u) = By(l — u)
The second section of the curve also has special blending functions.

By(u) = (1 — u)3/4
B3(u) = 7u°/12 — 5u7/4 + u/4 + 7/12
B3(u) = —u’/2 + u7/2 + u/2 + 1/6 ‘sles.
By(u) = w/6
And the next-to-the-last section of the curve uses the reverse of. these functions.

ta(u) = BA — u)
Bor(u) = B3(1 — u)
Byq(u) = BSC — u) we
Baa(u) = BY — u)
We can implement the B-spline curve-drawing method in the same way
as we did our
homemade blending functions. Instead of the SET-SMOOTH
routine, we need an al-
gorithm which will evaluate the five B-spline blending function groups.
This is done
by the SET-B-SPLINE routine. The order and form of calculation
of the B-spline func-
 CURVES AND FRACTALS 413

tions are such as to take advantage of expressions which appear more than once, to
avoid repeated computation.

11.10 Algorithm SET-B-SPLINE(NUMBER-OF-LINES) User routine to set the

number of line segments per curve section for B-spline interpolation
Argument NUMBER-OF-LINES the number of line segments per curve section
Global LINES-PER-SECTION storage for the NUMBER-OF-LINES value
BLEND, FIRST-BLEND, SECOND-BLEND, NEXT-TO-LAST-BLEND,
LAST-BLEND
4 by MAX-NUMBER-OF-LINES arrays to store blending function values
Local I, J to step through the needed curve points
U for blending function evaluation
Constant MAX-NUMBER-OF-LINES maximum line segments per curve section
BEGIN
IF NUMBER-OF-LINES < 1
OR NUMBER-OF-LINES > MAX-NUMBER-OF-LINES THEN
RETURN ERROR ‘INVALID NUMBER OF LINE SEGMENTS’;
LINES-PER-SECTION <- NUMBER-OF-LINES;
FOR I = 1TO NUMBER-OF-LINES DO
BEGIN
U <—1/ NUMBER-OF-LINES;
FIRST-BLEND[1, I] <—(1 — U) f 3;
FIRST-BLEND{4, I] —(U 7 3)/6;
FIRST-BLEND{3,I] <— (3/2 — 11 *U/ 12) *(U f 2);
FIRST-BLEND[2, I] <—1 — FIRST-BLEND(I, I] — FIRST-BLEND{3, J]
— FIRST-BLEND(4, I];
SECOND-BLEND[1, I] <— FIRST-BLEND[1, I] / 4;
SECOND-BLEND{4, I] < FIRST-BLEND[(4, IJ;
SECOND-BLEND{3, I] <— (((1 — U) « U + 1) * U + 1/3)/2;
SECOND-BLEND[2, I] <— 1 — SECOND-BLEND{(I, J]
— SECOND-BLEND{3, I] — SECOND-BLEND{4, I);
BLEND(1, I] < FIRST-BLEND[(I, I]/ 6;
BLEND(4, I] < FIRST-BLEND(4, Ij;
BLEND{3, I] < SECOND-BLEND{3, I];
BLEND(2, I] < 1 — BLEND[1, I] — BLEND(3, I] — BLEND(4, I);
J — NUMBER-OF-LINES — I;
IF] #0 THEN
BEGIN
NEXT-TO-LAST-BLEND[1, J] < SECOND-BLEND[(4, IJ;
NEXT-TO-LAST-BLEND[2, J] <— SECOND-BLEND{3, I];
NEXT.-TO-LAST-BLEND{3, J] < SECOND-BLEND(2, IJ;
NEXT-TO-LAST-BLEND{4, J] < SECOND-BLEND[I, J;
LAST-BLEND[I, J] < FIRST-BLEND{4, I);
LAST-BLEND[2, J] < FIRST-BLEND{3, I];
LAST-BLEND{3, J] <— FIRST-BLEND[2, I];
LAST-BLEND{4, J] <— FIRST-BLEND(I, IJ;
END;
END;
414 CHAPTER ELEVEN

NEXT-TO-LAST-BLEND[1, NUMBER-OF-LINES] <— 0;

NEXT-TO-LAST-BLEND[2, NUMBER-OF-LINES] < 1 / 6;
NEXT-TO-LAST-BLEND[3, NUMBER-OF-LINES] <— 7 / 12;
NEXT-TO-LAST-BLEND[4, NUMBER-OF-LINES] <1 /4;
LAST-BLEND[1, NUMBER-OF-LINES] < 0;
LAST-BLEND[2, NUMBER-OF-LINES] < 0;
LAST-BLEND[3, NUMBER-OF-LINES] < 0;
LAST-BLEND[4, NUMBER-OF-LINES] < 1;
RETURN;
END;

To start the B-spline curve, we require the first five sample points. As with our
first blending function, we start the curve by loading the sample points into the XSM,
YSM, ZSM arrays and calling the MAKE-CURVE routine with the special-starting
blending functions. The following algorithm starts the B-spline curve.

11.11 Algorithm START-B-SPLINE(AX, AY, AZ) User routine to start a B-spline

interpolation curve
Arguments AX, AY, AZ five-element arrays containing the first five points
Global XSM, YSM, ZSM four-element arrays for passing sample points
FIRST-BLEND, SECOND-BLEND
4 by LINES-PER-SECTION arrays containing blending function values
Local I for stepping through sample points
BEGIN
FOR I = 1TO4D0O
BEGIN
XSM[I] — AX[I);
YSM[I] <— AY[I];
ZSM[I] < AZ{I);
END;
MOVE-ABS-3(AX[1], AY[1], AZ[1]):
MAKE-CURVE[FIRST-BLEND];
NEXT-SECTION;
PUT-IN-SM(AX[5], AY[5], AZ[5]):
MAKE-CURVE[SECOND-BLEND];
NEXT-SECTION;
RETURN;
END;
Once the curve has been started, it can be continued by calling the CURVE-
ABS-3 routine. Each call of CURVE-ABS-3 adds another section to the curve.
To terminate the B-spline curve, we must draw the last two sections. This is done
by the END-B-SPLINE routine. This routine takes as arguments the last two sample
points for the curve. It processes these points using the NEXT-TO-LAST-BLEND and
LAST-BLEND blending function values.

11.12 Algorithm END-B-SPLINE(X1, Y1, Z1, X2, Y2, Z2) User routine to terminate
a B-spline curve
Arguments X1, Y1, Z1 the next-to-the-last sample point
X2, Y2, Z2 the last sample point
 CURVES AND FRACTALS 415

Global NEXT-TO-LAST-BLEND, LAST-BLEND

4 by LINES-PER-SECTION arrays of blending function values
BEGIN
PUT-IN-SM(X1, Y1, Z1);
MAKE-CURVE(NEXT-TO-LAST-BLEND);
NEXT-SECTION;
PUT-IN-SM(X2, Y2, Z2);
MAKE-CURVE(LAST-BLEND);
RETURN;
END;

The special B-spline blending functions used at the beginning (and end) of the
curve cause the curve to begin (end) at the first (last) sample point. So the curve will
be attached to the first and last sample points. Also, the slope of the curve at the end-
points is equal to the slope of the line segment connecting the first and second (next-
to-last and last) sample points.
The B-spline blending function can also be used to smooth polygons. To do so,
we need only call SET-B-SPLINE instead of SET-SMOOTH before calling the
SMOOTH-POLY-ABS-3 routine. (See Figure 11-9.)

B SPLINES AND CORNERS

The B-spline blending functions were designed to eliminate sharp corners in the curve,
and the curve does not usually pass through sample points. However, a sharp corner
and passage through a sample point can be produced in a B-spline curve, if desired.
The way that this is done is to use several identical sample points. The call

CURVE-ABS-3(X0, YO, ZO)

produces one sample point at (X0, YO, ZO), pulling the curve in that direction. (See
Figure 11-10a.)
The two calls

CURVE-ABS-3(X0, YO, ZO)

CURVE-ABS-3(X0, YO, ZO)

4
Pp we
|
/ \ Ze |
||
|
|
|
| FIGURE 11-9
EZ J Smoothing a polygon using B splines.
 (d)

FIGURE 11-10
(a) One point. (6) Two points. (c)
(c) Three points.

416
 CURVES AND FRACTALS 417

produce two sample points, both pulling the curve to the same place. The curve will be
pulled closer to this point, and the corner will bea little sharper. (See Figure 11-10b.)
If three identical sample points are used,

CURVE-ABS-3(X0, YO, ZO)

CURVE-ABS-3(X0, YO, ZO)
CURVE-ABS-3(X0, YO, ZO)

the curve will be forced through this point. (See Figure 11-10c.)
Blending-function interpolation and approximation by small straight-line seg-
ments allow us to draw curves which may not have simple mathematical forms. The
technique also lets us use the simple straight-line transformation and clipping methods
which we have developed. The major drawback of this technique is the amount of dis-
play-file storage required. For each sample point, 5 or 10 small line segments may be
produced, so care must be taken to assure sufficient display-file storage when curves
are drawn.

CURVED SURFACE PATCHES

We have seen a way of specifying curved lines with B splines. We use a set of sample
points or control points and the B-spline blending functions to determine points on the
curve. This method can be generalized to describe curved surfaces such as airplane
wings or car fenders. To describe a surface we shall need a grid of control points (x;;,
Yij> Zij)- (See Figure 11-11.)
Stepping through i moves one direction on the grid, while changing j moves in
the other direction. Such a grid can be used to define patches which will fit together to
form a smooth surface, just as we previously defined curve sections which fit together
to make the curve.
The coordinate values for points on the surface of a patch are given by

FIGURE 11-11
e A grid of control points.
418 CHAPTER ELEVEN

n m

i=1 j=l
k= fx(u, v) = > > x,;B,(u)B,(v)

n m
y=fyu,v)= = & y,Bi()B(v) (11.19)
i=l j=l

n m
z = fz(u; v) = > 2 Z,;B,(u)B;(v)
i=l j=l
There are two parameters which specify the position on the surface patch, u and v. Just
as for curves, each varies from 0 to i. Changing u will cause movement along the
patch in the same direction as changing i on the grid of control points. Changing v
moves in the same direction as changing j.
As with curves, we can move from one patch to the next by simply relabeling the
control points. For bicubic surface patches (cubic splines in both directions), the B
functions are those presented in Equations 11.14 through 11.18. The special functions
11.15 through 11.18 are used at the boundary of the overall surface. They should be
used for B; when the patch is first or second (last or next-to-last) in the i direction on
the grid, and similarly for B; and the j direction.
To represent a curved surface on a display, we must either rely on shading tech-
niques or show the curves which result from lines drawn on the surface. (See Figure
11-12.)
For example, for a given patch
x = fx(u, v), y = fy(u, v), z-= fz(u,,v) (11.20)

we can draw the curves resulting from fixing v at a constant and letting u vary from 0
to 1. We could draw the curves for several values of v. Similarly, we could draw curves
for fixed values of u, letting v change from 0 to 1. Notice that for this means of dis-
play, our B-spline curve-drawing algorithms suffice. If x for the surface is given by
n m
oe = x,B,(u)B,(v) (11.21)
i=l j=l

FIGURE 11-12
Displaying a curved surface patch.
 CURVES AND FRACTALS 419

and v is held constant at v,, then B,(v.) is a constant, so we can write

n m
ee [es X,;B,(v)]B;(u)
i=l j=l
(i192)
n
= = &B(u)
r="
where
m
X= ee X;jB\(Vo)
A=
(11.23)
This is just the expression used for the B-spline curve, provided the control point at
each i is replaced by the weighted sum over j of the grid points with index i. Thus for a
fixed v., we can find the values of B;(v.), perform the weighted sum of grid points in
the j direction to determine %; values, and use these as the control points in our curve-
drawing routines. Of course, the same thing can be done for the y and z coordinates.
If we wish to employ shading to display the curved surface, one approach is to
approximate the surface with a number of flat polygons. Each polygon can have a
slightly different direction and shade. We have already developed the machinery for
shading flat polygons, so the only thing new is the automatic decomposition of the
curved surface into a patchwork of flat surfaces. This is analogous to approximating a
curved path bya series of straight-line segments. One thought would be to find the
rectangular polygons formed by lines of constant u and v. But there is no guarantee that
four corner points of a small rectangular patch will all lie in the same plane (in fact,
they probably won’t). This may not bother us; if the patch is small, the points may be
nearly planar and the shading software may not require planar points. An alternative is
to decompose the surface into triangular pieces. Triangles are necessarily planar. Let’s
outline one method for converting to triangles. We have a unit square in the uv parame-
ter space which generates the surface patch. We first imagine the diagonal (0, 0) to
(1, 1) on this unit square. (See Figure 11-13.) This divides the square into two trian-
gles. The second step is to find a rule which divides a triangle into smaller triangles.
The rule is to find the midpoints of each of the three sides of the triangle and to con-

(0,1) (l, 1) LAV

(0, 0) (1, 0) | \
u
FIGURE 11-13
Form two triangles in uv parameter space.
420 CHAPTER ELEVEN

nect them. This divides the triangle into four smaller triangles. (See Figure 11-14.) We
can recursively apply this rule to break up the uv area into a mesh of small triangles.
Finally, we can use the vertices of the uv triangles in our B-spline surface equations to
find the corresponding xyz points. (See Figure 11-15.) These points form triangles
which can be shaded and entered into the display file.

BEZIER CURVES
Now let’s discuss a different curve construction called a Bezier curve. This is not so
much a different curve as it is a different way of specifying the curve. In fact, the same
shapes can be described by B splines and Bezier curves. We shall restrict our attention
to the cubic Bezier curve, which is adequate for most graphics applications. The cubic
Bezier curve requires four control points. These points completely specify the curve.
Unlike the B-spline curve, we cannot add additional points and smoothly extend the
cubic Bezier curve, but we can pick four more points and construct a second curve
which can be attached to the first. And if we are careful in our choice of the points, we
can make the second curve join the first curve smoothly.
A Bezier curve and its control points are shown in Figure 11-16. Note that the
curve begins at the first control point and ends on the fourth. This means that if we
want to connect two Bezier curves, we just make the first control point of the second
curve match the last control point of the first curve. Note also that at the start of the
curve, it is tangent to the line connecting the first and second control points. Likewise,
at the end of the curve, it is tangent to the line connecting the third and fourth control
points. That means that if we want to join two Bezier curves smoothly, we must ar-
range for the third and fourth control points of the first curve to be on the same line as
the first and second control points of the second curve.
Equations describing the Bezier curve are as follows:
X = xu? + 3x,0-(1 =u) + 3x oto) eat %y (lass
y = ygu’ + 3y3u7(1 — u) + 3y,u(1 — uw)? +y,(1 — 3 (11.24)
z= zqu° + 3z,u°(1 — u) + 3z,u(1 — u)* +z,(1 — uy
In these expressions, as u goes from 0 to 1, the curve goes from the first to the fourth
control point. (This is a bit different from our previous curves, where we only covered

FIGURE 11-14
Subdivide triangles by connecting midpoints.
 CURVES AND FRACTALS 42]

(0, 1) d, 1)

(0, 0) (1, 0)
u

FIGURE 11-15
Approximate the surface with a grid of triangles.

the region between two control points when the parameter changed from 0 to 1.) From
the expression, we could extract the blending functions for the Bezier curve and pro-
ceed as we did above; however, one of the most amazing things about Bezier curves is
that they can be constructed without reference to the above defining equations. In fact,
a Bezier curve can be constructed simply by taking midpoints. Consider Figure 11-17.
We have drawn the lines connecting the four control points A, B, C, and D. We have
determined the midpoints of these segments AB, BC, and CD. We connect these mid-
points with line segments and find their midpoints ABC and BCD. Finally, we connect
these two points and find the midpoint ABCD. Now the point ABCD is on the Bezier
curve. It divides the curve into two sections. Furthermore, the points A, AB, ABC,
and ABCD are the control points for the first section and the points ABCD, BCD, CD,
and D are the control points for the second section. Thus by taking midpoints, we can
find a point on the curve and also split the curve into two pieces complete with control
points. We can continue to split the curve into smaller sections until we have sections
so short that they can be replaced by straight lines, or even until the sections are no big-
ger than individual pixels.
Algorithms for adding Bezier curves to our system are given below. They assume
that the first control point is at the current pen position, so only three additional points
must be specified. They use the recursive subdivision approach.
422 CHAPTER ELEVEN

FIGURE 11-16
A cubic Bezier spline.

11.13 Algorithm BEZIER-ABS-3(XB, YB, ZB, XC, YC, ZC, XD, YD, ZD, N) Adds
a Bezier curve section to the display file
Arguments XB, YB, ZB, XC, YC, ZC, XD, YD, ZD control points
N the desired degree of subdivision
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
Local XAB, YAB, ZAB, XBC, YBC, ZBC, XCD, YCD, ZCD first-level mid-
points
XABC, YABC, ZABC, XBCD, YBCD, ZBCD second-level midpoints
XABCD, YABCD, ZABCD third-level midpoint
BEGIN
IFN = 0 THEN
BEGIN %
LINE-ABS-3(XB, YB, ZB);
LINE-ABS-3(XC, YC, ZC);
LINE-ABS-3(XD, YD, ZD);
END
ELSE
BEGIN
XAB < (DF-PEN-X + XB) / 2;
YAB < (DF-PEN-Y + YB) / 2;
ZAB < (DF-PEN-Z + ZB) / 2;
XBC < (XB + XC)/2;
 CURVES AND FRACTALS 423

YBC <— (YB + YC)/2;

ZBC <— (ZB + ZC) / 2;
XCD < (XC + XD) /2;
YCD <— (YC + YD)/2;
ZCD <— (ZC + ZD)/ 2;
XABC < (XAB + XBC)/ 2;
YABC < (YAB + YBC)/ 2;
ZABC <— (ZAB + ZBC)/ 2;
XBCD < (XBC + XCD) / 2;
YBCD <— (YBC + YCD) / 2;
ZBCD <— (ZBC + ZCD)/ 2;
XABCD < (XABC + XBCD)/ 2;
YABCD <— (YABC + YBCD) / 2;
ZABCD < (ZABC + ZBCD) / 2;
BEZIER-ABS-3(XAB, YAB, ZAB, XABC, YABC, ZABC, XABCD,
YABCD, ZABCD, N — 1);
BEZIER-ABS-3(XBCD, YBCD, ZBCD, XCD, YCD, ZCD, XD, YD, ZD,
Nod)
END;
RETURN;
END;

FIGURE 11-17
A Subdivision of a Bezier spline.
424 CHAPTER ELEVEN

Bezier curves satisfy the convex hull property. This means that the curve lies
within the boundary formed by an elastic band stretched over all four of the control
points. This property is useful in clipping tests.
Like B splines, the Bezier curyes can be extended to surfaces. We can construct
a bicubic Bezier surface from a4 X 4 grid of control points. Recursive subdivision can
be used to draw such patches by repeatedly quartering them until they may be replaced
by polygons or pixels.

FRACTALS
Man-made or artificial objects usually have either flat surfaces, which can be de-
scribed with polygons, or smooth curved surfaces, which we have just studied. But ob-
jects occurring in nature often have rough, jagged, random edges. Attempting to draw
things like mountains, trees, rivers, or lightning bolts directly with lines or polygons
requires a lot of specification. It is desirable to let the machine do the work of specify-
ing the jagged lines. We would like to give the machine two endpoints and let the
machine draw a jagged line between them (say for the path of a lightning bolt). Not
just any jagged line will do; we want a line which closely approximates the behavior of
nature, so it will look right. It is only recently that we have been able to do this be-
cause it is only recently that a branch of mathematics has been developed to describe
such rough structures. They are called fractals, and in this section we shall give an in-
troduction to what a fractal is and provide sample algorithms for generating them.
We begin with a discussion of dimension. Imagine that an object is composed of
elastic or clay. If the object can be deformed into aline or line segment, we give it di-
mension D, = 1. If it deforms into a plane or halfplane or a disk (as would a square),
we give it dimension D, = 2. And if it deforms into all space or half-space or into a
ball, we give it dimension D, = 3. We call D, the topological dimension.
Now let’s develop a second measure of an object’s dimension. Imagine taking a
line segment of length L and dividing it into N identical pieces, each of length / = L/N.
The pieces each look like the original, only scaled by a factor 1/s = ]/ L. To assemble
the original line segment from the segments scaled by 1/s, we must add together

N=s! (11.25)
of them. (Pretty easy so far; we just said if you divide a line into N pieces, it takes N
pieces to put it back together.) So next let’s consider a square. If we scale it down by
1/s, we will get a small square. How many small squares does it take to rebuild the
original? In the case of s = 2, it takes 4 squares, and for s =*3 it takes 9. In general
we have

N= 5" (11.26)
For a cube, if we scale by 1/s, we find that the number of small cubes needed to
assemble a large cube is

N=s° (11.27)
So we see a relationship. In going from line to square to cube, the exponent goes from
1 to 2 to 3. The exponent is measuring the dimension. So we can write the following
 CURVES AND FRACTALS 425

definition for dimension. If we scale an object by s and must assemble N of them to re-
construct the full-sized object, then the dimension D of the object is given by the rela-
tion

N = 8? (11.28)
We shall call D the fractal dimension. Solving for D gives
D = log N/logs (11.29)
Now let’s consider a rather strange construction call a Peano curve or space-fill-
ing curve. The particular Peano curve we shall focus on is called Hilbert’s curve. This
curve can be built by the following successive approximations. Begin with a square.
The first approximation will be to divide the square into 4 quadrants and draw the
curve which connects the center points of each. (See Figure 11-18.)
The second approximation will be to further subdivide each of the quadrants and
connect the centers of each of these finer divisions before moving to the next major
quadrant. (See Figure 11-19.)
The third approximation subdivides again. It again connects the centers of the
finest level before stepping to the next level of detail. (See Figure 11-20. )
Now we can imagine continuing this process indefinitely. What will the final re-
sult look like? For one thing, the curve never crosses itself. At each subdivision, the
curve fills smaller quadrants, but never crosses into an area where it already exists.
Another interesting fact is that the curve is arbitrarily close to every point in the
square. The curve passes through the points on a grid, which becomes twice as fine
with each subdivision, and there is no limit to the subdivisions. The curve fills the
square. One more fact is that the length of the curve is infinite. With each subdivision,
the length increases by a factor of 4. Since we imagine no limit to the subdivisions,
there is also no limit to the length. So we have constructed a curve which is topologi-
cally equivalent to a line D, = 1, but the curve has been so twisted and folded that it
exactly fills up a square. We can find the fractal dimension of the curve. At each sub-

FIGURE 11-18
The first approximation to Hilbert’s curve.
426 CHAPTER ELEVEN

FIGURE 11-19
The second approximation to Hilbert’s curve.

division the scale changes by 2, but the length changes by 4. Just as for the square, it
takes 4 curves of half scale to build the full-sized object, so the dimension D given by
4=2? (11.30)
must be D = 2. The Hilbert curve has topological dimension 1 but fractal dimension
2. It is a line so folded that it behaves like a two-dimensional object.
Now let’s look at one more strange curve called the triadic Koch curve. Begin
with a line segment. Divide it into thirds and replace the center third by the two adja-
cent sides of an equilateral triangle. (See Figure 11-21.) This gives us a curve which be-
gins and ends at the same place as the original segment but is built of 4 equal length
segments, each 1/3 the original length. So the new curve has 4/3 the length of the orig-
inal segment. Repeat the process for each of the 4 segments. (See Figure 11-22.) The
curve has gained more wiggles, and its length is now 16/9 times the original. Now

FIGURE 11-20
The third approximation to Hilbert’s curve.
 CURVES AND FRACTALS 427

imagine repeating the replacements indefinitely. Since each repetition increases the
length by a factor of 4/3, the length of the curve will be infinite, but it is folded in lots
of tiny wiggles. Unlike the Peano curve, it does not fill an area. In fact, it does not de-
viate very much from its original shape. Its topological dimension is 1, but what will
its fractal dimension be?
If we reduce the scale by 3, we find a curve that looks just like the original; but
we must assemble 4 such curves to make the original, so we have
4 = 3 (11.31)
Solving this for the fractal dimension gives
D = log34 = log 4/ log 3 = 1.2618 (11.32)
We get a nonintegral dimension! The curve acts like something between | and 2 dimen-
sions. This is all great fun, and it has led us to the definition of a fractal. Point sets,
curves, and surfaces which give a fractal dimension (a Hausdorff Besicovitch dimension,
to be precise) that is greater than the topological dimension are called fractals. The Hil-
bert curve and the Koch curve are fractals because their fractal dimensions (respec-
tively, 2 and 1.2618) are greater than their topological dimension of 1.
Nearly all the fractals which are of interest are self-similar under some change in
scale, either strictly, as in the case of Peano and Koch curves, or statistically. As an
example of a statistically self-similar fractal, consider a map of some coastline. The
coastline will have a number of wiggles, bays, and peninsulas. Now if we look more
carefully at the coastline, we will see a number of small variations which the map-
maker smoothed out. If we take a small section of the coast and map it more carefully
(using a finer scale), we will again see wiggles, bays, and peninsulas. Furthermore,
the shape and distribution of these wiggles will be the same as for the original map so
that we cannot tell just by looking at the map what scale was used. The locations of the
wiggles will, of course, be different, but their distribution (their likelihood) will be the
same. Coastlines are one example of fractals found in nature. They typically have frac-
tal dimension of about 1.15 to 1.25. There are many other examples. Bends and
branches in rivers, bursts of noise in transmission lines, clustering of stars, craters on
the moon, shapes of snowflakes, mountain landscapes, eddies and turbulence, branch-
ing of trees, lightning bolts, and many other phenomenon have been expressed as
fractals.

FRACTAL LINES
We can use the computer to easily generate self-similar fractal curves. The self-similar
drawing can be done by a self-referencing procedure (a recursive procedure). A curve

FIGURE 11-21
The replacement for a line segment for the triadic
Koch curve.
428 CHAPTER ELEVEN

FIGURE 11-22 .
The second approximation to the Koch curve.

which is composed of N self-similar pieces, each scaled by 1/s, can be drawn by a

routine which calls itself N times with arguments scaled by 1/s. Of course, a computer
routine should terminate, which a true fractal does not. In the computer routine, each
recursive call has smaller arguments, that is, smaller lengths. There will be some point
where the lengths become smaller than the size of a pixel. There is not much point in
continuing the recursion beyond this since the wiggles will be smaller than a pixel and
cannot be displayed. So the computer procedures can terminate when lengths become
less than a pixel and still provide the computer’s best approximation to the fractal.
Being able to use the computer to generate fractal curves means that the user can
easily generate realistic coastlines or mountain peaks or lightning bolts without con-
cern for all the small bends and wiggles. The computer can generate the wiggles, and
the user need only give the endpoints. So let’s write a simple fractal line-drawing al-
gorithm fora statistically self-similar curve. (See Figure 11-23.) Start by looking at the

FIGURE 11-23
A fractal line.
 CURVES AND FRACTALS 429

halfway point for the fractal line. If we had a perfectly straight line between (X), Yq, Z1)
and (X2, Y2, Z) the halfway point would be the midpoint

(* Ae OR ie eee hee *) (11.33)

? >
2 2 2
But for our fractal line, the halfway point will probably be offset alittle from this, sy
we add an offset term to each coordinate. (See Figure 11-24.)

Ki he X: et Zipaez
(SEEsoe MEM ay, <2 + ax) (11.34)

In order to get the random effect, the amount of the offset should be chosen at
random. The offset is calculated as

dx = L* W* GAUSS (11.35)
Identical expressions give dy and dz. The variable L is the length of the segment. The
W is a weighting function governing the curve roughness (and fractal dimension). The
function GAUSS returns a Gaussian variable with 0 mean. That means that (a) about
half the values returned are positive and the other half are negative and (b) numbers
greatly different than 0 are much less likely to appear than those near 0. The Gaussian
distribution describes the way things often cluster around an average (such as the way
people’s height varies from average height). The pseudo-random number function
(RND) available with many programming languages is usually a uniform distribution;
any number between 0 and | is equally likely to occur. This is not what we need, but
an approximate Gaussian distribution can be derived from it as shown in algorithm
11.16. The offset calculation says that the halfway point for the fractal line differs from
the ideal midpoint according to the laws of chance. The scale factor tells how strongly
the line tends to deviate. So far, we just have a line with a corner in it, not a full fractal
line, but to further bend it we just recursively apply our method to find halfway points
for each subsegment. The recursion continues until the segments become too small to
matter.
We can see that this results in a fractal curve because it can be divided into two
pieces (N = 2), each piece being statistically similar to the original and having a
length averaging a bit greater than half the original length. The dimension will be abit
greater than 1. The greater W, the rougher the curve; and the greater the length, the
smaller the value of s by which it must be scaled to give the original. We see that in-
creasing W will increase the fractal dimension.
Algorithm 11.14 presents a procedure for drawing a fractal line segment from the
current position to the specified point. It requires as arguments the endpoint, a weight

FIGURE 11-24
Offset the halfway point.
430 CHAPTER ELEVEN

factor, and the desired recursion depth. An alternative implementation might determine
the appropriate depth of recursion automatically by determining when the midpoint
offset is probably less than a pixel. To be really correct, we should multiply the weight
W by the length of each segment for each recursive call; but for the sake of efficiency,
we shall perform the multiplication only once at the start (and with an approximation
to the length) and then assume that at each level the length is close to half the previous
length so that we can just divide by 2.

11.14 Algorithm FRACTAL-LINE-ABS-3 (X, Y, Z, W, N, FSEED) User routine for

drawing fractal lines
Arguments X, Y, Z the point to which to draw the line
W describes the roughness of the curve
N the desired depth of recursion
FSEED seed for fractal pattern
Global DF-PEN-X, DF-PEN-Y, DF-PEN-Z current pen position
SEED the seed used by the random number generator
Local L the approximate line length
BEGIN
SEED < FSEED;
L<|X — DF-PEN-X| + |Y — DF-PEN-Y| + |Z — DF-PEN-Z |;
FRACTAL-SUBDIVIDE(DF-PEN-X, DF-PEN-Y, DF-PEN-Z, X, Y, Z, L * W, N);
END;

The SEED is a number given to the random number generator. It is assumed that
RND not only returns a random number calculated from SEED but also alters the value
of SEED so that on the next call a different random number will be returned. The
reason for giving the SEED value as an argument is that the particular shape of the
fractal will depend upon the initial seed used. This means that if we wished to draw
two similar fractals at different positions (for example, to represent the edges of each
half of a torn piece of paper), we could do so by calling our fractal-drawing routine
twice, with arguments the same except for the start and endpoints.
The routine which actually does most of the work is the following:

11.15 Algorithm FRACTAL-SUBDIVIDE(X1, Y1, Z1, X2, Y2, Z2, S, N) Draws a

fractal line between points X1, Y1, Z1 and X2, Y2, Z2
Arguments X1, Y1, Z1 the point to start the line
X2, Y2, Z2 the point to stop the line Nx
S offset scale factor ‘.
N the desired depth of recursion
Local XMID, YMID, ZMID coordinates at which to break the line
BEGIN
IF N = 0 THEN
BEGIN
recursion stops, so just draw the line segment
LINE-ABS-3(X2, Y2, Z2)
END
ELSE
 CURVES AND FRACTALS 431

BEGIN
calculate the halfway point
XMID < (X1 + X2)/2 + S x GAUSS;
YMID <—(Y1 + Y2)/2 + S * GAUSS;
YMID <— (Y1 + Y2)/2 + S * GAUSS;
draw the two halves
FRACTAL-SUBDIVIDE(X1, Y1, Z1, XMID, YMID, ZMID, S/2, N—1);
FRACTAL-SUBDIVIDE(XMID, YMID, ZMID, X2, Y2, Z2, S/2, N—1);
END;
RETURN;
END;

We approximate a Gaussian distribution by averaging several uniformly random

numbers. Half the numbers are added and half subtracted to provide for zero mean.

11.16 Algorithm GAUSS Calculates an approximate Gaussian distribution between — 1

and 1.
Local I for summing samples
BEGIN
GAUSS < 0;
FOR I = 1TO 6 DO GAUSS «+ GAUSS + RND — RND;
GAUSS < GAUSS / 6;
RETURN;
END;

As in the case of curves, we have implemented the FRACTAL-LINE-ABS-3 al-

gorithm so that it breaks up the fractal into straight-line segments and then places them
into the display file. The advantage of this is that all of the clipping and transformation
machinery can be applied. The disadvantage is that a large display file is needed to
hold all of the tiny line segments, and time must be spent processing each of them. For
some applications it may be better to process the fractal when it is removed from the
display file and not deal with all of the pieces until they are actually displayed. If this
approach were to be taken, then the seed values for the fractal would have to be stored
as part of the display-file entry.

FRACTAL SURFACES
A fractal line is fine for the path of a lightning bolt, but for something like a three-di-
mensional mountain range, we needafractal surface. There are several ways to extend
the fractal idea to surfaces. The one we shall present is based on triangles. Given three
vertex points in space, we shall generate a fractal surface for the area between them.
There are methods for decomposing arbitrary polygons into triangles, so the method
can be used to cover more general shapes. The idea is as follows: Consider each edge
of the triangle. We can imagine afractal line along each edge and compute its halfway
point by the same means as we used for fractal lines. (See Figure 11-25.)
Now by connecting these halfway points with line segments, we can subdivide
the surface into four smaller triangles. (See Figure 11-26.) We can then recursively
 FIGURE 11-25
Find the halfway point (with offset) for each of the
triangle sides.

FIGURE 11-26
Subdivide the triangle into smaller triangles.

432
 CURVES AND FRACTALS 433

FIGURE 11-27
A fractal surface.

apply the method to each of the small triangles to subdivide the surface even further.
We can continue this subdivision until the triangles are too small to matter. (See Figure
11-27.)
The algorithm presented is not very efficient in that it calculates each point twice
(once for each of the triangles which border on it). Care must be taken with the seed
values to ensure that the same fractal edge is generated for the two bordering triangles.

11.17 Algorithm FRACTAL-SURFACE(AX, AY, AZ, ASEED, W, N) Creates a fractal

surface fromatriangle
Arguments AX, AY, AZ three-element arrays containing the triangle vertices
434 CHAPTER ELEVEN

ASEED, athree-element array of seeds for the fractal patterns of the three
triangle sides
W fractal roughness weight
N recursion depth
Local L1, L2, L3 for the approximate lengths of the triangle sides
BEGIN
L1 = |AX[(2] — AX[(1]| + |AY[2] — AY[1]| + |AZ[2] — AZ[1]];
L2 = |AX[3] — AX(2]| + |AY[3] — AY[2]| + |AZ[3] — AZ[2]|;
L3 = |AX[1] — AX[3]| + |AY[1] — AY[3]] + |AZ[1] — AZ[3]|;
FRACTAL-SURFACE-SUBDIVIDE(AX[1], AY[1], AZ[1], ASEED[1], L1 * W,
AX[2], AY[2], AZ[2], ASEED[2], L2 « W,
AX[3], AY[3], AZ[3], ASEED[3], L3 « W, N, 1);
RETURN;
END;

This is just a setup routine; the actual subdivision work is done in algorithm
FRACTAL-SURFACE-SUBDIVIDE.

11.18 Algorithm FRACTAL-SURFACE-SUBDIVIDE(X1, Y1, Z1, S1, W1, X2, Y2,

22, S2, W2, X3, Y3, Z3, S3, W3, N, DIR) Generates a fractal triangle
Arguments X1, Y1, Zl, X2, Y2, Z2, X3, Y3, Z3 the three vertices of the triangle
S1, S2, S3 seed values for the fractal patterns of the three sides
W1, W2, W3 roughness weights times the lengths
N recursion depth
DIR direction indicator for drawing the triangle
Global SEED the random number generator seed
DF-PEN-X, DF-PEN-Y, DF-PEN-Z the current pen position
Local X12, Y12, Z12, X23, Y23, Z23, X31, Y31, Z31 coordinates of the division
points
BEGIN
IFN = 0 THEN
BEGIN
recursion stops, so draw the triangle
DF-PEN-X < X3;
DF-PEN-Y < Y3;
DF-PEN-Z <— Z3;
DISPLAY-FILE-ENTER(3);
IF DIR > 0 THEN 44
BEGIN
LINE-ABS-3(X1, Y1, Z1);
LINE-ABS-3(X2, Y2, Z2);
LINE-ABS-3(X3, Y3, Z3);
END
ELSE
BEGIN
LINE-ABS-3(X2, Y2, Z2);
LINE-ABS-3(X1, Y1, Z1);
LINE-ABS-3(X3, Y3, Z3);
END;
 CURVES AND FRACTALS 435

END
ELSE
BEGIN
calculate the halfway points
SEED < S1;
X12 — (X1 + X2)/2 + W1 * GAUSS;
Y12 —(Y1 + Y2)/2 + W1 « GAUSS;
Z12 —(Z1 + Z2)/2 + W1 * GAUSS;
SEED <— 82;
X23 < (X2 + X3)/2 + W2 * GAUSS;
Y23 <— (¥2 + Y3)/2 + W2 * GAUSS;
Z23 < (Z2 + Z3)/2 + W2 * GAUSS;
SEED < §3;
X31 <— (X3 + X1)/2 + W3 * GAUSS;
Y31 —(Y3 + Y1)/2 + W3 * GAUSS;
Z31 <—(Z3 + Z1)/2 + W3 * GAUSS;
draw the four subtriangles
FRACTAL-SURFACE-SUBDIVIDE(X1, Y1, Z1, S1+ 1, W1/2,
XA GO Zilla Wels
ASL YS, Z31, S343; W3/2)
N-—i, DIR);
FRACTAL-SURFACE-SUBDIVIDE(X2, Y2, Z2, S2+ 1, W2/2,
OS 23 L250 521 oe Wole,
2 V2 ZS Sts, WiI/2*
N-1, DIR);
FRACTAL-SURFACE-SUBDIVIDE(X3, Y3, Z3, S3 + 1, W3/2,
X31, Y31, Z31, S3+2, W1/2,
X23 WY235 223,-S2 +3, W2/2,
N-—1, DIR);
FRACTAL-SURFACE-SUBDIVIDE(X12, Y12, Z12, S1+2, W2/2
XS ERYSIY 2314 S32 awil/2s
X23) 23223, S2+2,.W3/2,
N= = DIR):
END;
RETURN;
END;

We have presented algorithms for drawing statistically self-similar fractal curves

and surfaces, but these are by no means the only ways to construct such figures. Fur-
thermore, there are countless fractal structures with exact self-similarity, many of
which produce striking images.

AN APPLICATION
Let’s consider the drawing of smooth curves once again. An area in which curve draw-
ing is useful is the design of objects such as airplane wings or automobile fenders. Let
us outline how our curve-drawing routines might be used in a program for the interac-
tive design of a curved line such as a wing cross section. We shall provide the designer
with 10 sample points. He will be able to select a sample point with the pick, and
436 CHAPTER ELEVEN

move it with the locator. The B-spline routines will be used to interpolate the sample
points. As a sample point is moved, the curve will change. The designer arranges the
sample points until he obtains the curve which he desires.
Our first problem will be to initialize the sample points. We shall use the charac-
r ‘‘X’’ to display the points. The points will be constructed at the origin, and the
image transformation will be used to position them on the screen.

BEGIN
FORI = 1TO 10 DO
BEGIN
CREATE-SEGMENT(I);
MOVE-ABS-2(0, 0);
IUD.GNE.Cyp
CLOSE-SEGMENT;
SET-VISIBILITY(I, TRUE);
SET-DETECTABILITY(I, TRUE);
SET-IMAGE-TRANSLATION(I, I/ 10, I/ 10);
END;
SET-B-SPLINE(8);
RETURN;
END;

Once this initialization is done, we shall have 10 points displayed along the
screen diagonal. (See Figure 11-28.)
We can now enter the main part of the program, which consists of a loop allow-
ing repeated modification and display of the curve. The part of the program for moving
a sample point might look like the following:

BEGIN
SAMPLE <0;
WHILE SAMPLE = 0 DO AWAIT-PICK(WAIT, SAMPLE);
BUTTON-NUM < 0;
WHILE BUTTON-NUM = 0 DO
AWAIT-BUTTON-GET-LOCATOR(WAIT, BUTTON-NUM, X, Y);
SET-IMAGE-TRANSLATION(SAMPLE, X, Y);
END;

FIGURE 11-28
Ten sample points along the diagonal.
 CURVES AND FRACTALS 437

Now we need some statements which will draw the curve using the positions of
the sample points. We shall need some arrays to hold the sample points, so suppose
that AX, AY, and AZ are five-element arrays and that AZ contains zero values. We
shall get the first five sample-point positions and place them into the arrays for use by
START-B-SPLINE. The next three points will be handled by CURVE-ABS-3, while
the last two points are needed for END-B-SPLINE. The curve will be drawn in dis-
play-file segment 11. (See Figure 11-29.)

BEGIN
DELETE-SEGMENT(11);
CREATE-SEGMENT(11);
FOR I = 1 TO 5 DO INQUIRE-IMAGE-TRANSLATION(I, AX[I], AY[I]);
START-B-SPLINE(AX, AY, AZ);
FOR I = 6TO 8 DO
BEGIN
INQUIRE-IMAGE-TRANSLATION(I, X, Y);
CURVE-ABS-3(X, Y, 0);
END;
INQUIRE-IMAGE-TRANSLATION(9, X, Y);
INQUIRE-IMAGE-TRANSLATION(10, U, V);
END-B-SPLINE(X, Y, 0, U, V, 0);
CLOSE-SEGMENT;
SET-VISIBILITY(11, TRUE);
SET-DETECTABILITY(11, FALSE);
END;

FURTHER READING
An introduction to ways of representing curves may be found in [FOR69]. Circle
generators are described in [BRE77], [DOR79], [HOR76], [MCI83], and [SUE79]}.
Conic generators are given in [PIT67], [PRO83], and [SMI71]. Curve generators can
be divided into those based on a parametric form x = fx(u,v), y = fy(u,v), and those
based on an implicit form f(x,y) = 0. Curve generators for the implicit form are de-
scribed in [COH76], [DAN70], [JOR73 ], [VAN84a], and [VAN85]. Conversion be-
tween the parametric and implicit forms is presented in [SED84] and [SED86]. Some
texts with deeper discussions of curves are [CHA78], [DEB78], and [ROG76]. Conic
curves are discussed in [COH76], [PAV83], [PIE84], and [PRA85]. Splines are sur-
veyed in [BAR84] and [BOE82]. Bezier curves are described in [GOR74]. The sub-

FIGURE 11-29
Positioning the sample points to produce the
x desired shape curve.
438 CHAPTER ELEVEN

division of cubic Bezier and B-spline curves is presented in [LAN80b]. Spline tech-
niques can be applied to all four homogeneous coordinates instead of just three, as we
have done. The result is called a rational spline. Rational cubic splines can represent
both cubic curves and conic sections. A description may be found in [AHU68] and
[TIL83]. Representing surfaces by a number of patches is described in [FOR72]. Fit-
ting data sample points with splines is discussed in [PLA83] and [YAM78]. The prob-
lem of finding the B-spline control points which will describe a surface passing
through a grid of sample points is considered in [BAR80]. It is possible to subdivide B
splines in a manner similar to that for Bezier curves [COH80]. Replacing curves with
line segments is considered in [COH69], [CRA85], and [LER 69]. Several techniques
have been used to image curved surfaces [FOR79]. One method is to subdivide the sur-
face into polygons or pixels. See [CAT75], [LAN79], and [LAN80a]. They have also
been imaged directly [SCH82]. Displaying contour lines for the surface is described in
[SAT85]. Handling hidden curved surfaces is considered in [GRI84]. Ray tracing can
be used to display curved objects, provided we can find the intersection of a ray with
an object; methods for doing this for bicubic patches are discussed in [KAJ82] and
[SWE86]. Ray tracing of objects defined by sweeping or revolving curves is discussed
in [KAJ83a] and [KAJ83b] and, for curves that are cubic splines, in [VAN84b]. Ray
tracing of parametric surfaces is discussed in [TOT85]. Mapping textures onto curved
surfaces is described in [BLI76]. Programs for drawing Hilbert’s curve are given in
[ALE71], [GOL81], [GRI85], and [MOR82]. Fractals are defined in [MAN83] and
[MAN77]. Some simple fractal curve generators are presented in [CAR86]. The statis-
tical fractals we have presented are described in [FOU82]. Some objections to this sim-
ple approach were raised in,[MAN82]. A different approach to constructing fractals
based on iterated function systems is given in [BAR85] and [DEM85]. Ray-tracing
techniques can be applied to fractal surfaces, provided we can find the intersection be-
tween the ray and the surface. One approach described in [BOU85] is to determine if
the ray intersects a bounding ellipsoid; if so, the fractal is subdivided and the test is
reapplied. Using a ‘‘cheesecake-shaped’’ box to bound the fractal is described in
[KAJ83a] and [KAJ83b]. The idea of describing objects by statistical models has been
extended to volumes of particles, which can yield impressive results (see Plates 8 and
9) for trees, grass, clouds, and flames [REE85].

{[AHU68] Ahuja, D. V., “‘An Algorithm for Generating Spline-Like Curves,’ JBM System Journal, vol. 7,
nos. 3&4, pp. 206-215 (1968). 4
[ALE71] Aleph Null, ‘‘Space Filling Curves, or How to Waste Time with a Rlotter.”’ Software—Practice and
Experience, vol. 1, no. 4, pp. 403-410 (1971).
[BAR80] Barsky, B. A., Greenberg, D. P., ‘‘Determining a Set of B-Spline Control Vertices to Generate
an
Interpolating Surface,’’ Computer Graphics and Image Processing, vol. 14, no. 3, pp. 203-226
(1980).
[BAR84] Barsky, B. A., ‘‘A Description and Evaluation of Various 3-D Models,’’ JEEE Computer
Graphics
and Applications, vol. 4, no. 1, pp. 38-52 (1984).
[BAR85] Barnsley, M. F, Demko, S., ‘‘Iterated Functions Systems and the Global Construction
of Frac-
tals,’ Proceedings of the Royal Society of London, Series A, vol. 399, no. 1817, pp. 243-276 (1985).
{BLI76] Blinn, J. R, Newell, M. E., ‘‘Texture and Reflection in Computer Generated Images,’’
Communi-
cations of the ACM, vol. 19, no. 10, pp. 542-547 (1976).
 CURVES AND FRACTALS 439

[BOE82] Boehm, W., ‘‘On Cubics: A Survey,”’ Computer Graphics and Image Processing, vol. 19, no.
3,
pp. 201-226 (1982).
[BOU85] Bouville, C., ‘‘Bounding Ellipsoids for Ray-Fractal Intersection,”’ Computer Graphics, vol. 19,
no. 3, pp. 45-52 (1985).
[BRE77] Bresenham, J., ‘‘A Linear Algorithm for Incremental Digital Display of Circular Arcs,’’ Communi-
cations of the ACM, vol. 20, no. 2, pp. 100-106 (1977).
[CAR86] Carlson, P. W., ‘‘IBM Fractal Graphics,’’ Compute!, vol. 8, no 3, pp. 78-80 (1986).
(CAT75] Catmull, E., ‘“Computer Display of Curved Surfaces,’’ IEEE Conference on Computer Graphics,
Pattern Recognition, and Data Structures, pp. 11-17, IEEE Cat. No. 75CH0981-1c (1975).
[CHA78] Chasen, S. H., Geometric Principles and Procedures for Computer Graphic Application, Prentice-
Hall, Englewood Cliffs, N.J. (1978).
[COH69] Cohen, D., Lee, T. M. P., ‘‘Fast Drawing of Curves for Computer Display,’’ AFIPS SJCC, vol.
34, pp. 297-307 (1969).
[COH76] Cohen, E., “‘A Method for Plotting Curves Defined by Implicit Equations,” Computer Graphics,
vol. 10, no. 3, pp. 263-265 (1976).
[COH80] Cohen, E., Lyche, T., Risenfeld, R., ‘‘Discrete B-Splines and Subdivision Techniques in Com-
puter-Aided Geometric Design and Computer Graphics,’’ Computer Graphics and Image Processing,
vol. 14, no. 2, pp. 87-111 (1980).
[CRA85] Crampin, M., Guifo, R., Read, G. A., ‘‘Linear Approximation of Curves with Bounded Curva-
ture and a Data Reduction Algorithm,’’ Computer-Aided Design, vol. 17, no. 6, pp. 257-261 (1985).
[DAN70] Danielsson, P. E., ‘‘Incremental Curve Generation,’’ JEEE Transactions on Computers, vol. C-19,
no. 9, pp. 783-793 (1970).
[DEB78] de Boor, C., A Practical Guide to Splines, Applied Mathematical Sciences Series, vol. 27, Springer-
Verlag, New York (1978).
[DEM85] Demko, S., Hodges, L., Naylor, B., ‘‘Construction of Fractal Objects with Iterated Function Sys-
tems,’’ Computer Graphics, vol. 19, no. 3, pp. 271-278 (1985).
[DOR79] Doros, M. ‘‘Algorithms for Generation of Discrete Circles, Rings, and Disks,’’ Computer
Graphics and Image Processing, vol. 10, no. 4, pp. 366-371 (1979).
[FOR69] Forrest, A. R., “‘Curves for Computer Graphics’’ in Pertinent Concepts in Computer Graphics,
Univ. of Ill. Press (1969), pp. 31-47.
[FOR72] Forrest, A. R., “On Coons and Other Methods for the Representation of Curved Surfaces,’’ Com-
puter Graphics and Image Processing, vol. 1, no. 4, pp. 341-359 (1972).
[FOR79] Forrest, A. R., ‘‘On the Rendering of Surfaces,’’ Computer Graphics, vol. 13, no. 2, pp. 253-259
(1979).
{FOU82] Fournier, A., Russell, D., “Computer Rendering of Stochastic Models,’’ Communications of the
ACM, vol. 25, no. 6, pp. 371-384 (1982).
[GOL81] Goldschlager, L. M., “‘Short Algorithms for Space-Filling Curves,’’ Software—Practice and Ex-
perience, vol. 11, no. 1, pp. 99-100 (1981).
[GOR74] Gordon, W. J., Riesenfeld, R. F, ‘‘Bernstein-Bezier Methods for the Computer-Aided Design of
“Free-Form Curves and Surfaces,’’ Journal of the ACM, vol. 21, no. 2, pp. 293-310 (1974).
[GRI84] Griffiths, J. G., ‘‘A Depth-Coherence Scanline Algorithm for Displaying Curved Surfaces,’’ Com-
puter-Aided Design, vol. 16, no. 2, pp. 91-101 (1984).
[GRI85] Griffiths, J. G., ‘“Table-Driven Algorithms for Generating Space-Filling Curves,’’ Computer-Aided
Design, vol. 17, no. 1, pp. 37-41 (1985).
[HOR76] Horn, B. K. P., “‘Circle Generators for Display Devices,’’ Computer Graphics and Image Process-
ing, vol. 5, no. 2, pp. 280-288 (1976).
[JOR73] Jordon, B. W., Lennon, W. J, Holm, B. D., ‘‘An Improved Algorithm for the Generation of Non-
parametric Curves,’’ IEEE Transactions on Computers, vol. C-22, no. 12, pp. 1052-1060 (1973).
[KAJ82] Kajiya, J. T., ‘‘Ray Tracing Parametric Patches,’’ Computer Graphics, vol. 16, no. 3, pp. 245-254
1982).
Firs cae J. T., ‘‘New Techniques for Ray Tracing Procedurally Defined Objects,’’ Computer
Graphics, vol. 17, no. 3, pp. 91-102 (1983).
[KAJ83b] Kajiya, J. T., ‘‘New Techniques for Ray Tracing Procedurally Defined Objects,’’ ACM Transac-
tions on Graphics, vol. 2, no. 3, pp. 161-181 (1983).
440 CHAPTER ELEVEN

[LAN79] Lane, J., Carpenter, L., ‘‘A Generalized Scan Line Algorithm for the Computer Display of
Parametrically Defined Surfaces,’’ Computer Graphics and Imaging Processing, vol. 11, no. 3, pp.
290-297 (1979).
[LAN80qa] Lane, J. M., Carpenter, L. C., Whitted, T., Blinn, J. F, ‘* Scan Line Methods for Displaying
Parametrically Defined Surfaces,’’ Communications of the ACM, val. 23, no. 1, pp. 23-34 (1980).
[LAN80b] Lane, J. M., Riesenfeld, R. FE, ‘‘A Theoretical Development for the Computer Generation and
Display of Piecewise Polynomial Surfaces,’’ JEEE Transactions on Pattern Analysis and Machine Intel-
ligence, vol. PAMI-2, no. 1, pp. 35-46 (1980).
[LER69] Le Riche, P. J., ‘‘A Curve Plotting Procedure,’’ Computer Journal, vol. 12, pp. 291-292 (1969).
[MAN77] Mandelbrot, B. B., Fractals—Form, Chance, and Dimension, W. H. Freeman, San Francisco
(1977).
[MAN82] Mandelbrot, B. B., ‘‘Comment on Computer Rendering of Fractal Stochastic Models,’’ Com-
munications of the ACM, vol. 25, no. 8, pp. 581-584 (1982).
[MAN83] Mandelbrot, B. B., The Fractal Geometry of Nature, W. H. Freeman, San Francisco (1982).
[MCI83] McIlroy, M. D., “‘Best Approximate Circles on Integer Grids,’’ ACM Transactions on Graphics,
vol. 2, no. 4, pp. 237-263 (1983).
[MOR82] Morrison, R., ‘‘Low Cost Computer Graphics for Micro Computers,’’ Software—Practice and Ex-
perience, vol. 12, no. 8, pp. 767-776 (1982).
[PAV83] Pavlidis, T., ‘‘Curve Fitting with Conic Splines,’’ ACM Transactions on Graphics, vol. 2, no. 1, pp.
1-31 (1983).
[PIE84] Piegl, L., ‘‘Defining C1 Curves Containing Conic Segments,’’ Computers & Graphics, vol. 8, no.
2, pp. 177-182 (1984).
{PIT67] Pitteway, M. L. V., “‘Algorithm for Drawing Ellipses or Hyperbolae with a Digital Plotter,’ Com-
puter Journal, vol. 10, no. 3, pp. 282-289 (1967).
[PLA83] Plass, M., Stone, M., ‘‘Curve-Fitting with Piecewise Parametric Cubics,’’ Computer Graphics,
vol. 17, no. 3, pp. 229-239 (1983).
[PRA85] Pratt, V., ““Techniques for Conic Splines,’’ Computer Graphics, vol. 19, no. 3, pp. 151-159
(1985).
[PRO83] Prosser, C. J., Kilgour, A. C., ‘‘An Integer Method for the Graphical Output of Conic Sections,”’
ACM Transactions on Graphics, vol. 2, no. 3, pp. 182-191 (1983).
[REE85] Reeves, W. T, Blau, R., ‘“‘Approximate and Probabilistic Algorithms for Shading and Rendering
Structured Particle Systems,’’ Computer Graphics, vol. 19, no. 3, pp. 313-322 (1985).
[ROG76] Rogers, D. FE, Adams, J. A., Mathematical Elements for Computer Graphics, McGraw-Hill, New
York (1976).
[SAT85] Satterfield, S. G., Rogers, D. F,, ‘‘A Procedure for Generating Contour Lines from a B-Spline Sur-
face,’’ IEEE Computer Graphics and Applications, vol: 5, no. 4, pp. 71-75 (1985).
[SCH82] Schweitzer, D., Cobb, E. S., ‘‘ Scanline Rendering of Parametric Surfaces,’’ Computer Graphics,
vol. 16, no. 3, pp. 265-271 (1982).
[SED84] Sederberg, T. W., Anderson, D. C., ‘‘Implicit Representation of Parametric Curves and Surfaces,”’
Computer Graphics and Image Processing, vol. 28, no. 1, pp. 72-84 (1984).
[SED86] Sederberg, T. W., Goldman, R. N., ‘‘Algebraic Geometry for Computer-Aided Geometric De-
sign,’’ Computer Graphics and Applications, vol. 6, no. 6, pp. 52-59 (1986).
[SMI71] Smith, L. B., ‘‘Drawing Ellipses, Hyperbolas, or Parabolas with a Fixed Number of Points and
Maximum Inscribed Area,’’ Computer Journal, vol. 14, no. 1, pp. 81-86 (1971).
[SUE79] Suenaga, U. Y., Kame, T., Kobayashi, T., ‘‘A High-Speed Algorithm for the Generations of
ae Lines and Circular Arcs,’’ IEEE Transactions on Computers, vol. C-28, no. 10, pp. 728-736
1979).
[SWE86] Sweeney, M. A. J., Bartels, R. H., ‘“‘Ray Tracing Free-Form B-Spline Surfaces,’’ Computer
Graphics and Applications, vol. 6, no. 2, pp. 41-49 (1986).
[TIL83] Tiler, W., ‘‘Rational B-Splines for Curve and Surface Representation,’’ IEEE Computer Graphics
and Applications, vol. 3, no. 9, pp. 61-69 (1983).
[TOT85] Toth, D. L., ‘‘On Ray Tracing Parametric Surfaces,”’ Computer Graphics, vol. 19,
no. 3, pp. 171-
179 (1985).
[VAN84q] Van Aken, J. R., “‘An Efficient Ellipse-Drawing Algorithm,’ IEEE Computer Graphics and Appli-
cations, vol. 4, no. 9, pp. 24-35 (1984).
 CURVES AND FRACTALS 44]

[VAN846] Van Wijk, J. J., ‘‘Ray Tracing Objects Defined by Sweeping Planar Cubic Splines,’’ ACM Trans-
actions on Graphics, vol. 3, no. 3, pp. 223-237 (1984).
[VAN85] Van Aken, J., Novak, M., ‘‘Curve-Drawing Algorithms for Raster Displays,’ ACM Transactions
on Graphics, vol. 4, no. 2, pp. 147-169 (1985).
[YAM78] Yamaguchi, F, “‘A New Curve Fitting Method Using a CRT Computer Display, ’’ Computer
Graphics and Image Processing, vol. 7, no. 3, pp. 425-437 (1978).

EXERCISES
11-1 Circular arcs may be supported by hardware drawing primitives in the display. If we are to
take advantage of this feature, we need a means of allowing the user to specify an arc and also a
way of describing a circular arc in our display file.
a) How many reasonable ways of specifying an arc can you devise?
b) Create a display-file instruction for a circular arc such that we can easily perform
rotation, translation, and uniform scaling operations.
c) Choose a representation which also preserves the proper starting and ending positions
of the arc on all transformations.
11-2 Design a circle-drawing algorithm which does antialiasing.
11-3 a) Calculate the values of our homemade blending functions B,, B, B3, B, needed when
each curve section is to be approximated by five line segments.
b) Sketch these blending functions.
c) If the four sample points (1, 1, 0), (2, 1, 0), (2, 2, 0), (1, 2, 0) are used, calculate and
plot the interpolated curve for the center section.
11-4 a) Calculate the values of the cubic B-spline blending functions for the first, second, and
middle curve sections (B, B’, B; for i = 1, 2, 3, 4) when each curve section is to be approxi-
mated by three line segments.
b) Sketch these blending functions.
c) If the first six sample points are (1, 1, 0), (2, 1, 0), (2, 2, 0), (1, 2, 0), ©, SO) SO Be
0), calculate and plot the first three sections of the interpolated curve.
11-5 Show that the four B-spline blending functions in Equation 11.14 sum to 1. What does this
mean?
11-6 Afractal curve named the dragon is constructed by repeatedly replacing each line segment
by two line segments forming a right angle.

What is the fractal dimension of this curve?

442 CHAPTER ELEVEN

PROGRAMMING PROBLEMS
11-1 a) Implement algorithm 11.1 for drawing arcs.
b) Extend your graphics system to include arc-drawing instructions as investigated in
Exercise 11-1. ‘ .
c) Develop and implement a clipping algorithm for circular arcs.
11-2 Suppose you wish to display the points on some function y = f(x) which you can calcu-
late. One way to do this is to calculate sample points along the function and then to connect
them with line segments (as we did in Programming Problem 1-6). Write a recursive procedure
which will do this, subdividing the curve with each call. If possible, let the function to be plot-
ted be an argument to your procedure. At each recursive call you must check to see if the curve
is sufficiently subdivided to be approximated bya straight line. Try the following three criteria
for this test.
a) A specified number of levels has been reached.
b) The length of the approximating line segment is less than some value.
c) The angle between segments is less than some value.
Try plotting functions which vary slowly and functions with sharp corners, noting the quality of
the image and the number of line segments required.
11-3 a) Implement algorithms 11.2 through 11.9 for Lagrange interpolation.
b) Also implement algorithms 11.10 through 11.12 for B splines.
11-4 Test the Lagrange and B-spline interpolation by constructing a set of sample points. Show
the sample points as follows:
a) Connected by single line segments
b) Interpolated using the Lagrange interpolation scheme with eight line segments per
curve section
c) Interpolated using the cubic B splines with eight line segments per curve section
11-5 Use the Lagrange interpolation routines to present a smooth curve of the widget sales data
of Programming Problem 1-3.
11-6 Sketch a face and use B splines to smooth the features.
11-7 Write an interactive graphing program which draws an approximate curve fora set of data
points and then allows the user to adjust the curve for the desired fit.
11-8 Show the effect of smoothing on the regular polygons with three to eight sides. Show each
figure both before and after smoothing.
11-9 Write strictly two-dimensional versions of the B-spline routines.
11-10 Implement and test algorithm 11.13 for drawing Bezier curves. Also implement relative
and strictly two-dimensional versions of the procedure.
11-11 Write procedures for entering and displaying bicubic Bezier surface patches.
11-12 Implement algorithms 11.14 through 11.18 for fractal lines and surfaces.
11-13 a) Write a program which generates Hilbert’s curve to any desired approximation.
b) Write one for the triadic Koch curve.
11-14 Write the routines FRACTAL-LINE-ABS-2, FRACTAL-LINE-REL-2, and FRACTAL-
LINE-REL-3 to extend the system to absolute and relative fractal lines in both two
and three di-
mensions.
11-15 Generalize the fractal surface-drawing routine to arbitrary convex polygons
by subdi-
viding the polygons into triangles and then applying the FRACTAL-SURFACE-SUBDIVIDE
routine.
11-16 In our graphics system, the procedures which enter a polygon into the display
have avail-
able all of the polygon’s vertices and know just how many sides the polygon
will have. An alter-
 CURVES AND FRACTALS 443

native used by some graphics systems is to construct a path, a sequence of line segments, and
then, at some arbitrary point, close the path to form a polygon. Polygons are built a side at a
time and need not be specified all at once. This method easily generalizes to allow both straight
and curved path elements, so polygons can have a mixture of straight and curved sides. Write
procedures BEGIN-PATH and CLOSE-PATH which give this feature to our system. BEGIN-
PATH should divert MOVE and LINE drawing display-file commands to a temporary holding
area. CLOSE-PATH should transfer the collected commands to the display file as a polygon.
APPENDIX

INFORMATION
GUIDE

There is an enormous wealth of literature about computer graphics; just listing it would
fill a book by itself. In this appendix we shall provide some direction on how and
where to find this literature.

GUIDES
Machover, C., ‘‘Background and Source Information about Computer Graphics,’” JEEE Computer Graphics
and Applications, vol. 5, no. 1, pp. 68-81 (1985).
Machover, C., ‘‘An Updated Guide to Sources of Information about Computer Graphics,’’ JEEE Computer
Graphics and Applications, vol. 3, no. 1, pp. 49-S9 (1983).
Machover, C., ‘‘A Guide to Sources of Information about Computer Graphics,’’ IEEE Computer Graphics
and Applications, vol. 1, no. 1, pp. 73-85 (1981).
Schrack, G. FE, ‘‘Gaining Access to the Literature of Computer Graphics,’’ JEEE Transactions on Education,
vol. E-26, no. 1, pp. 7-10 (1983).

BIBLIOGRAPHIES
Bibliographies cataloging each year’s literature have been produced. They cover the
years from 1975 to date. There are also occasional bibliographies for some subfields
and related areas. The following is a list of the major ones.

Schrack, G. F, “‘Literature in Computer Graphics for the Year 1984: A Bibliography,’ Computer Graphics,
vol. 19, no. 2, pp. 41-77 (1985).
Schrack, G. F, “Literature in Computer Graphics for the Year 1983: A Bibliography,’’ Computer Graphics,
vol. 18, no. 2, pp. 69-89 (1984).
Schrack, G. F,, “‘Literature in Computer Graphics for the Year 1982: A Bibliography,’’ Computer Graphics,
vol. 17, no. 4, pp. 173-204 (1983).

444
 INFORMATION GUIDE 445

Schrack, G. F, ‘‘Literature in Computer Graphics for the Year 1981: A Bibliography,’’ Computer Graphics,
vol. 17, no. 2, pp. 115-141 (1983).
Schrack, G. F, ‘‘Current Literature in Computer Graphics and Interactive Techniques: References, Eighth
Series,’’ Computer Graphics, vol. 16, no. 4, pp. 251-262 (1982).
Schrack, G. E, ‘‘Computer Graphics: A Keyword-Indexed Bibliography for the Year 1980,’’ Computer
Graphics and Image Processing, vol. 18, no. 2, pp. 145-187 (1982).
Schrack, G. F, ‘‘Current Literature in Computer Graphics and Interactive Techniques: References, Seventh
Series,’’ Computer Graphics, vol. 16, no. 1, pp. 101-118 (1982).
Schrack, G. F, ‘‘Current Literature in Computer Graphics and Interactive Techniques: References, Sixth
Series,’’ Computer Graphics, vol. 15, no. 4, pp. 458-464 (1981).
Schrack, G. FE, “‘Computer Graphics: A Keyword-Indexed Bibliography for the Year 1979,’’ Computer
Graphics and Image Processing, vol. 15, no. 1, pp. 45-78 (1981).
Schrack, G. F, “‘Current Literature in Computer Graphics and Interactive Techniques: References, Fifth
Series,’’ Computer Graphics, vol. 14, no. 4, pp. 200-218 (1980).
Schrack, G. F, ‘‘Current Literature in Computer Graphics and Interactive Techniques: References, Fourth
Series,’’ Computer Graphics, vol. 16, nos. 1 & 2, pp. 73-83 (1980).
Schrack, G. F, ‘‘Current Literature in Computer Graphics and Interactive Techniques: References,’’ Com-
puter Graphics, vol. 13, no. 4, pp. 381-394 (1980).
Schrack, G. FE, ‘‘Computer Graphics: A Keyword-Indexed Bibliography for the Years 1976, 1977, and
1978,’ Computer Graphics and Image Processing, vol. 14, no. 1, pp. 24-79 (1980).
Schrack, G. F, ‘‘Current Literature in Computer Graphics and Interactive Techniques,’’ Computer Graphics,
vol. 13, no. 1, pp. 148-165 (1979).
Schrack, G. FE, ‘‘Current Literature in Computer Graphics and Interactive Techniques,’’ Computer Graphics,
vol. 12, no. 4, pp. 114-123 (1978).
Pooch, U. W., ‘‘Computer Graphics Interactive Techniques, and Image Processing 1970-1975: A
Bibliography,’’ Computer, vol. 9, no. 8, pp. 46-64 (1976).

Some special topic bibliographies are as follows:

Barsky, B. A., ‘‘Computer-Aided Geometric Design: A Bibliography with Keywords and Classified Index,”’
Computer Graphics, vol. 16, no. 1 (1982).
Griffiths, J.G., ‘‘Bibliography of Hidden-Line and Hidden-Surface Algorithms,’’ Computer Aided Design,
vol. 10, no. 3 (1978).
Rosenfeld, A., ‘‘Picture Rocessing: 1972,’’ Computer Graphics and Image Processing, vol. 1, no. 4, pp.
394-416 (1972).
Thalmann, N., Thalmann, D., ‘‘An Indexed Bibliography on Computer Animation,’’ [EEE Computer
Graphics and Applications, vol. 5, no. 7, pp. 76-86 (1985).
Thomas, J. J., Hamlin, G. ‘‘Graphical Input Interaction Technique (GIIT),’’ Computer Graphics, vol. 17,
mo. 1, pp. 5-30 (1983).

An overview of computer graphics and selected references may be found in

Burchi, R.S., ‘‘Interactive Graphics Today,”’’ JBM Systems Journal, vol 19, no. 3, pp. 299-313 (1980).

BOOK REVIEWS
Reviews of computer graphics texts may be found in the following:

Schrack, G. F, ‘‘Texts and Books in Computer Graphics,’’ Computer Graphics, vol. 18, no. 4, pp. 160-165
(1984).
Schrack, G. FE, ‘‘Book Reviews,’’ Computer Graphics, vol. 19, no. 2, pp. 77-82 (1985).
vol. 17,
Schrack, G. FE, ‘‘A Brief Survey of Texts and Books in Computer Graphics,’’ Computer Graphics,
no. 4, pp. 90-95 (1984).
446 apPENDIXA

TEXTS
Some of the currently available texts are as follows:

Angell, I. O., A Practical Introduction to Computer Graphics, Macmillan, London and Basingstoke, England
(1981). :
Chasen, S. H., Geometric Principles and Procedures for Computer Graphics Applications, Prentice-Hall, En-
glewood Cliffs, N.J. (1978).
Demel, J. T., and Miller, M. J., Introduction to Computer Graphics, Brooks/Cole, Belmont, Calif. (1984).
Encarnacao, J., and Schlectendahl, E. G., Computer Aided Design Fundamentals and System Architectures,
Springer-Verlag, New York (1983).
Enderle, G., Kansy, K., and Pfaff, G., Computer Graphics Programming GKS—The Graphics Standard,
Springer-Verlag, New York (1983).
Foley, J. D., and Van Dam, A., Fundamentals of Interactive Computer Graphics, Addison-Wesley, Reading,
Mass. (1981).
Newman, W., and Sproull, R., Principles of Interactive Computer Graphics (2d ed.), McGraw-Hill, New
York (1979).
Pavlidis, T., Algorithms for Graphics and Image Processing, Computer Science Press, Rockville, Md.
(1982).
Rogers, D. F, and Adams, J. A., Mathematical Elements for Computer Graphics, McGraw-Hill, New York
(1976).
Scott, J. E., Introduction to Interactive Computer Graphics, Wiley Interscience, New York (1982).
Sherr, S., Video and Digital Electronic Displays: A User’s Guide, Wiley, Somerset, N.J. (1982).
 APPENDIX

PIDGIN
ALGOL

The algorithms presented in this book have been written in an imaginary ALGOL-like
programming language. This language was chosen for several reasons. It allows us to
concentrate on the ‘‘meat’’ of the algorithm without too much concern over the details
of declarations and syntax. The actual implementation of the algorithms necessitates a
translation to some real programming language, so that a bit more study and thought
are needed than that required for simple copying. Translation to PASCAL, PL/I, C, and
FORTRAN can be carried out without much difficulty, and no single language is fa-
vored.
In this appendix we present a description of PIDGIN ALGOL to aid in un-
derstanding the algorithms which use it.

Comments. Comments may be entered at any point within the algorithm. They are dis-
tinguished by being typed in lowercase letters.
Identifiers. Names for variables and procedures consist of any number of letters, digits,
and hyphens, beginning with aletter.
Variables. A variable may be either an argument, a local, or a global. Local variables
exist only for the life of the routine. They do not retain their values over separate
calls. They are not known outside the routine. Global variables can be accessed
by any routine. They exist throughout the life of the program. Each global vari-
able has a unique name by which it is identified. Arguments are assumed bound

447
448 APPENDIX B

by a call by a reference mechanism, so that alterations in the value of the formal

argument of the called routine will cause a corresponding change in the as-
sociated variable of the calling routine. No data types for variables are specified.
Arrays. Arrays of data elements may be used. Individual array elements are specified
by the array name, followed byalist of indices enclosed in brackets A[I, J, K].
There is no restriction as to the number of allowed dimensions. There is also no
mechanism for declaring the size of the array. An entire array may be passed be-
tween routines by using the array name (without subscripts) as the argument.
Constants. Constants may be associated with identifiers. These identifiers may be used
in place of the constants to improve the comprehensibility of the routine.
Operators. The available arithmetic operators are as follows:

{? exponentiation
* multiplication
/ division
+ addition
— subtraction and negation

Operators must be separated from other symbols by at least one space.

The relational operators are

equal
not equal
less than
greater than
less or equal
I
VAVA*
greater or equal

The result of a relational operation is a value of TRUE or FALSE.

The logical operators are

NOT negation
AND conjunction
OR _ disjunction

A few useful functions will also be assumed to be present in the language. They
include
 PIDGINALGOL 449

INT(X) greatest integer part of X

MAX(X,Y) _ largest of X and Y
MIN(X,Y) smallest of X and Y
|x| absolute value of X
SIN(X) sine of X
COS(X) cosine of X
ARCTAN(X) arctangent of X
SQRT(X) square root of X

Statements. Routines are composed of statements, where each statement causes some
action to be performed. There can be several statements on one line, or a single
statement may require several lines. Statements are terminated by a semicolon.

<Statement 1>; <Statement 2>; <Statement 3>; <Statement 4>;

Compound Statements. Several statements may be grouped together so as to behave as a

single statement. This is done by enclosing these statements between the words
BEGIN and END.

BEGIN
<Statement |1>;
<Statement 2>;
<Statement 3>;
END;

Each routine consists of one compound statement.

Assignments. Variables may be assigned values by use of the assignment statement.

<Variable> < <Expression>;

as in

xX <—3;

The assignment operator is <— .

forms
Control Statements. Decisions are carried out by the IF statement, which has two
450 APPENDIX B

IF <Expression> THEN <Statement>;

and .

IF <Expression> THEN <Statement 1> ELSE <Statement 2>,;

The expression must evaluate to TRUE or FALSE. If the expression is TRUE,

then for the first form the statement will be executed, and for the second form,
<Statement 1> will be carried out. If the expression is FALSE, then nothing is
done in the first form, and <Statement 2> is executed for the second form. In
the event of nested IF statements, the ELSE clause is associated with the most re-
cent incomplete IF Thus in

IF <E1> THEN IF <E2> THEN <S1> ELSE <S2>;

the statement <S2> belongs with the test on <E2>.

The FOR construct allows a finite number of repetitions of a statement with vary-
ing index or counter. Its form is

FOR <Index> = <Start> TO <Finish> DO <Statement>:

With each repetition, the index is incremented by 1. The first time the <State-
ment> is executed, the index will have the value <Start>. Repetition continues
until the value of index exceeds <Finish>. If <Finish> is less than <Start>
then the index is decremented. The expressions <Start> and <Finish> are
evaluated only once upon entry of the loop.
The WHILE construct has the form

WHILE <Expression> DO <Statement>; -

The <Statement> is repeated until the <Expression> becomes FALSE. If the

<Expression> is FALSE upon entry to this construct, the <Statement> will not
be executed.
Functions and Procedures. Each routine is considered to be a separate subprogram,
callable from any other subprogram. The ordering of the various subprograms is
 PIDGINALGOL 45]

arbitrary. Each routine is headed by the routine’s name, followed byalist of its
formal arguments enclosed in parentheses.

Algorithm FOO(X, Y, Z)

Functions are treated the same as procedures with the exception that at some
point within the function body, the function name must be assigned the value
which is to be returned.

Algorithm THREE(Y)
BEGIN
THREE < 3 « Y;
END;

There is a RETURN; statement which may be used at any point to transfer con-
trol back to the calling routine. There is also a RETURN ERROR <String>;
construct which will print an error message containing <String> before
transferring control back to the calling routine.
A call of a procedure is a statement composed of only the procedure name and ar-
gument list.

FOO(2, 4, 6);

Functions are called by using them in expressions.

“7 <2 * THREE(4) + 7:

Input/Output Statements. Input may be done by the statement

READ <Variable>;

where <Variable> is the variable identifier to be given values. Output is carried

out by the statement
452 APPENDIX B

PRINT <Expression 1>, <Expression 2>, ... <Expression N>;

where <Expression 1>, ... <Expression N> are the.expressions whose values
are to be printed. Each print command utilizes a new output record (a new line).
Printing of a number of array elements may be conducted by the construction

PRINT FOR <Index> = <Start> TO <Finish>, <Array>[ <Index> ];

 APPENDIX

.
GRAPHICS
ON AN
ALPHANUMERIC
TERMINAL

Throughout this book we have stated that graphics output may be obtained on an ordi-
nary CRT terminal or line printer. In this appendix we present FORTRAN subroutines
which have been used to interface our graphics system to a Decwriter. CRT terminals
can be handled ina similar fashion. The routines differ from those for some line print-
ers in that the Decwriter routines do not require any initial carriage control character.
The routines display an image 90 characters wide by 50 lines high. Different characters
are used to obtain different line and interior styles.
Algorithm 1.2 BRESENHAM should be included among these device-dependent
interface routines. It will not be given here, but follows the form given in Chapter 1. A
90 x 50 element array IFRAME is used as the frame buffer.
Algorithm 1.3 ERASE clears the frame buffer array by setting each element to
the space character.

SUBROUTINE ERASE
COMMON /SBLK/ IWEND, IHEND,IWSTRT,IHSTRT,IWIDTH,
IHIGHT
COMMON /FBLK/ IFRAME(90,50)
DATA IBLNK /” ’/

453
454 AppENDIXC

Ee STEP THRU ARRAY SETTING ALL ELEMENTS TO BLANK

DO 120I = IWSTRT, IWEND
DO 110J = IHSTRT, IHEND
IFRAME(,J) = IBLNK
110 CONTINUE
120 CONTINUE
RETURN
END

Algorithm 1.4 DISPLAY prints out the contents of the frame buffer array in order
of top to bottom. Some blank lines are also printed to position one frame per page.

SUBROUTINE DISPLY
THIS ROUTINE DISPLAYS THE CONTENTS OF THE FRAME BUFFER
COMMON /SBLK/ IWEND, JHEND, IWSTRT, IHSTRT, WIDTH, IHIGHT
COMMON /FBLK/ IFRAME(90,50)
IHEND1 = IHEND + 1
DISPLAY THE ARRAY
WRITE (6,123)
123 FORMAT (///////)
DO 220 I = IHSTRT, IHEND
WRITE (6,210) (FRAME(J,IHEND1 —1), J=IWSTRT,IWEND)
210 FORMAT(90A1)
220 CONTINUE
WRITE (6,123)
RETURN
END

Algorithm 1.5 INITIALIZE-1 sets the display size parameters, and since car-
riage control is not available, the user is requested to position the paper.

SUBROUTINE INITL1
COMMON /SBL IHEND,
K/IWIWSTRT, IHSTRT,IWIDTH
END, , IHIGHT
IWEND = 90
IHEND = 50
IWSTRT = 1
IHSTRT = 1
IWIDTH = IWEND — IWSTRT wt
IHIGHT = IHEND — IHSTRT %
WRITE (6,10) >
10 FORMAT(‘ POSITION ATTOP OF PAGE, PUSH RETURN’)
READ (5,20) IDUMMY
20 FORMAT(A2)
RETURN
END

Algorithm 1.6 TERMINATE can be a do-nothing dummy routine (a STOP state-

ment might be included in this routine).
 GRAPHICS ON AN ALPHANUMERIC TERMINAL 455

SUBROUTINE TRMNAT
RETURN
END

Algorithm 2.8 DOMOVE needs only to update the variables used for the current
pen position.

SUBROUTINE DOMOVE(X, Y)
Cc THIS ROUTINE MOVES THE CURRENT PEN POSITION
COMMON /PDBLK/ PXAT, PYAT
COMMON /SBLK/ IWEND, IHEND, IWSTRT, IHSTRT,IWIDTH,
IHIGHT
PXAT = AMAX1(FLOAT(IWSTRT),AMIN1(FLOAT(IWEND),X*
1 IWIDTH + IWSTRT))
PYAT = AMAX1(FLOAT(IHSTRT),AMIN1(FLOAT(IHEND),Y*
1 IHiGHT + IHSTRT))
RETURN
END

Algorithm 2.9 DOLINE causes aline to be entered into the frame buffer array by
a call on the BRESENHAM routine. LINCHR is the current line style character.

SUBROUTINE DOLINE(X, Y)
Cc THIS ENTERS A LINE INTO THE FRAME BUFFER
COMMON /CBLK/ LINCHR
COMMON /SBLK/ IWEND, IHEND, IWSTRT,IHSTRT,IWIDTH,IHIGHT
COMMON /PDBLK/ PXAT, PYAT
X1 = PXAT
Y1 = PYAT
PXAT = AMAX1(FLOAT(IWSTRT),AMIN1(FLOAT(IWEND),X*
1 IWIDTH + IWSTRT))
PYAT = AMAX1(FLOAT(IHSTRT),AMIN1(FLOAT(IHEND),Y*
1 IHIGHT + IHSTRT))
CALL BRSNHM(INT(X1 + 0.5), INT(Y1 + 0.5),
INT(PXAT + 0.5), INT(PYAT + 0.5), LINCHR)
: RETURN
END

Algorithm 2.18 DOCHAR places a character into the frame buffer. The KODE
routine converts from ASCII character code to a form suitable for printing.

SUBROUTINE DOCHAR(IOP,X, Y)
C THIS ROUTINE ENTERS A CHARACTER INTO THE FRAME BUFFER
COMMON /SBLK/ IWEND,IHEND,IWSTRT,IHSTR T,
WIDTH, IHIGHT
COMMON /FBLK/ IFRAME(90, 50)
ICXAT = MAXO(IWSTRT,MINO(IWEND, INT(X*IWIDTH + IWSTRT)))
ICYAT = MAXO(IHSTRT, MINO(WHEND, INT(Y*IHIGHT + IHSTRT)))
ICHR = KODE(IOP)
456 APPENDIX C

IFRAME(ICXAT, ICYAT) = ICHR

RETURN
END
’

The initializations of algorithms 2.13, 2.19, and 2.24 can be combined into the
following subroutine.

SUBROUTINE INITL2
COMMON /CSZBLK/ CHARHT,CHARWD,CHRSEP
COMMON /DFBLK/ IFREE
COMMON /SBLK/ IWEND,IHEND,IWSTRT,IHSTRT,
[WIDTH IHIGHT
COMMON /PINBLK/ XAT, YAT
CALL INITL1
IFREE = 1
CALL DOSTYL(O)
CHARWD = 1.0 /FLOAT(IWIDTH)
CHARHT = 1.0 / FLOAT(IHIGHT)
CHRSEP = 0.0
CALL CHARUP(0.0,1.0)
XAT = 0.0
YAT = 0.0
CALL NEWFRM
RETURN
END

Algorithm 3.5 DOSTYLE changes the line or interior style by changing the
character used in the drawing. The default line style is an asterisk (*).

SUBROUTINE DOSTYL(IOP)
COMMON /PLYBLK/ IPLYCH
COMMON /CBLK/ LINCHR
IF (IOP .LT. — 15) GOTO 1228
LCHR = 42 — IOP
IF (LCHR .LE. 48) GO TO 1227
LCHR = LCHR - 16
1227 LINCHR = KODE(LCHR)
RETURN ae
1228 LCHR = 26-IOP *
IF (LCHR .LE. 48) GO TO 1229 7
LCHR = LCHR - 16
1229 IPLYCH = KODE(LCHR)
RETURN
END

The FILLIN routine called in algorithm 3.14 uses the BRESENHAM algorithm
call to fill in a scan line.
 GRAPHICS ON AN ALPHANUMERIC TERMINAL 457

SUBROUTINE FILLIN(X1, Y1,X2,Y2)

COMMON /PLYBLK/ IPLYCH
IF (X1 .EQ. X2) GOTO 1301
CALL BRSNHM(INT(X1 + 0.5), INT(Y1 + 0.5),
1 INT(X2 + 0.5), INT(Y2 + 0.5), IPLYCH)
1301 RETURN
END
 a| & inate AIH
SHG ng ee

nae
et —

Counce
£ Aetna enicriiiamn cael
COMO RIL: REE de Sk eae
Che IE Tw eS nr HES
oes - wie aR, WARS ?
; teea eg, See i aut
CM pOSTYL: | ae
Je cSt a PAW. On FLaS OTE, ees ma
: HAMET » fats BOARS Mn. a4
. Geesey & RE OR: . :
thd 5 CUD AREARREL DB
wae a 0.8 : ‘ ee!
— ats “Ze :
CLL WEWrRM = ed e
aE? es y DS , F ll 7 ; i ’

END ;

SUP RTE DOSE Yaa)

CAME PLY? TRACE
FHARON MRT SIND a
ae
08 Be Ee
La. ~$ Oe PO Cane 2 7
LEO Pa OOP
ee Ae EG re
13 HC
hare
as em

aed
 INDEX

Absolute line command, 38 B spline, 411, 438

Absolute move command, 38 Back clipping plane, 286
Absolute polygon command, 73 Back face, 312
Actual device coordinates, 41 BACK-FACE-CHECK, 318, 329
ADD-EVENT-Q, 215 Back-face removal, 312-318
ADD-STRING-Q, 216 BEZIER-ABS-3, 422
ADJUST-ORIGINAL-POLY, 336 Bezier curves, 420, 437
ALGOL, PIDGIN, 447-452 Binary space partition algorithm, 327, 341
Algorithm, 451 Bit block-transfer (BITBLT), 164, 166
Aliasing, 21, 26, 136, 140 Bitmap characters, 22
Angle of incidence, 348 Blending functions, 402-403, 412
Angle of reflection, 350 Boxing test, 322
Animation, 139, 166, 392 BRESENHAM, 18
Antialiasing: Bresenham’s algorithm, 17-20, 26
characters, 24, 26 Brush, 233
lines, 20-21, 26 BUILD-TRANSFORMATION, 128, 134,
patterns, 140 158
polygons, 94, 96, 106 Button, 212
Approximation curve, 400
Arc, 398 Cabinet projection, 278
ARCDDA, 339 CAD (computer-aided design), 341
Arrays, 448 Calligraphic displays, 36
ASCII character codes, 49-50 Cathode ray tube (CRT), 3, 34, 35
Assignment statement, 449 Cavalier projection, 278
Attenuation of light, 365 Center of curvature, 398
Attributes, 146 Center of projection, 264
AWAIT-BUTTON, 220 CGI (computer graphics interface), 2
AWAIT-BUTTON-GET-LOCATOR, 231 CGM (computer graphics metafile), 2
AWAIT-EVENT, 216 Character codes, 49-50
AWAIT-PICK, 221 Character generation, 22-24
Character-generator chip, 23
Axonometric projection, 276
459
460 InDeEx

Character positioning, 50-51 Coordinates (Cont. ):

CHECK-STYLE, 340 right-handed, 245
Chromaticity, 378 transformation, 118, 133-134
CIE (Commission Internationale view plane, 268, 279-282
L’Eclairage), 378 . COPY-LIGHT, 361
Circle generation, 398, 437 COPY-SEGMENT, 171
CLIP, 194, 298 CORE, 2, 26
CLIP-BACK, 291 Cosine, 112-113
CLIP-BOTTOM, 187, 295 CREATE-SEGMENT, 149, 180, 223, 300
CLIP-FRONT, 292 Cross hairs, 206
CLIP-LEFT, 186, 293 Cross product, 313
CLIP-POLYGON-EDGE, 192, 297, 316 CRT (cathode ray tube), 3, 34, 35
CLIP-RIGHT, 187, 294 Cubic B spline, 411
CLIP-TOP, 188, 296 Current position, 38
Clipping, 172, 181, 200, 304 Cursor, 206, 207
Cohen-Sutherland Outcode algorithm, CURVE-ABS-3, 408
182-183, 200 Curve generation algorithm, 397, 437
generalized, 196 Cylindrical coordinates, 383
planes, 290
polygons, 190-195 DASHED-ABS-2, 66
Sutherland-Hodgman algorithm, 184, DASHED-REL-2, 66
200 DDA (digital differential analyzer), 13, 26
three-dimensional, 285—298 DELETE-ALL-SEGMENTS, 153
volume, 285 DELETE-SEGMENT, 152
CLOSE-SEGMENT, 150, 328 Detectability, 222
CMY color model, 382 : Determinant, 124-125
Coefficient of reflection, 346 Device coordinates, 41
Cohen-Sutherland Outcode algorithm, Diffuse illumination, 345—347
182-183, 200 Digital differential analyzer (DDA), 13, 26
Color, 376 Digitizer, 207
Color models, 379-388, 392 Dimetric projection, 278
Color table, 389, 392 Direct memory access, 34
Comments, 447 Direct-view storage tube (DVST), 34-35
COMPARE-POLYS, 333 Direction number, 75
Compound statement, 449 DISABLE-ALL, 214
Computer-aided design (CAD), 341 DISABLE-GROUP, 213
Computer graphics interface (CGI), 2 DISPLAY, 25, 454
Computer graphics metafile (CGM), 2 Display control, 47—49
Concave polygons, 71 Display devices, 34.37
Constants, 448 Display file, 36,-42
Contour edge, 326 algorithms, 45—47
Control statement, 449 interpreter, 40-41, 45, 60
Convex hull, 424 segments, 146, 166
Convex polygons, 71 Structures, 161-163
Coordinates, 3, 244-245 DISPLAY-FILE-ENTER, 44, 57, 195,
cylindrical, 383 300, 355
homogeneous, 116, 252, 304 Display procedures, 136-140
left-handed, 245 Display processor, 40, 60
object, 269 Distance between point and line, 8-9
 INDEX 461

Distance between point and plane, 247 Front clipping plane, 286
Dither matrix, 374 Function, 450
DO-TRANSFORMATION, 127
DOCHAR, 54, 455 Gamma correction, 389
DOLINE, 46, 455 Gamut of colors, 378
DOMOVE, 45, 455 GAUSS, 431
DOPOLYGON, 79 General form of a line, 4
DOSTYLE, 56, 78, 456 Geometrically sorting, 319-320
Dot-matrix characters, 22, 26 GET-KEYBOARD-DATA, 219
Dot-matrix printer, 37 GET-PICK-DATA, 220
Dot product, 247, 315 GET-POINT, 43
Double buffering, 153 GET-TRANSFORMED-POINT, 131
Dragging, 237 GETQ, 218
DVST (direct-view storage tube), 34-35 GKS (graphics kernel system), 2, 26
Gloss, 350
Echoing, 230 Gouraud shading, 364
Edges, 71 Graph, 57-60, 94-96
Electrostatic plotter, 37 Graphics kernel system (GKS), 2, 26
ENABLE-GROUP, 212 Graphics standards, 2
END-B-SPLINE, 414 Graphics Standards Planning Committee of
END-CURVE, 408 the ACM, 2
ERASE, 24, 453 Grid, 233
Even-odd inside test, 75
EVENT, 214 Halftone, 373, 391
Event-driven device, 209 Hidden lines, 68—69
Event handling, 214-221 HIDDEN-SURFACE-CHECK, 331
Event queue, 211, 214, 240 Hilbert’s curve, 425, 438
EXTEND-SEGMENT, 171 HLS color model, 387
Homogeneous coordinates, 116, 252, 304
Facets, 350 HSI color model, 383
Feedback, 206 HSV color model, 385
FILL-POLYGON, 82 HSV-TO-RGB, 386
FILL-SCAN, 89 Hue, 378, 383, 392
FILLIN, 89, 92, 135, 456
Film printer, 36-37 Identifiers, 447
Flicker, 36 IDENTITY-MATRIX, 125
Flight simulation, 302, 305 Illumination:
Flood fill, 79, 96 diffuse, 345-347
FLUSH-ALL-EVENTS, 219 point source, 348
Fractal, 424, 438 Image space, 173
Fractal dimension, 425 Image transformation, 155-157
FRACTAL-LINE-ABS-3, 430 Incidence, angle of, 348
FRACTAL-SUBDIVIDE, 430 INCLUDE, 87
FRACTAL-SURFACE, 433 Index of refraction, 353, 365
FRACTAL-SURFACE-SUBDIVIDE, 434 INITIALIZE-1, 25, 454
Frame buffer, 11, 24, 60 INITIALIZE-2, 57, 456
Franklin algorithm, 325 INITIALIZE-2A, 49
Frequency, 376 INITIALIZE-2B, 54
Fresnel equation, 353 INITIALIZE-3, 93
462 invex

INITIALIZE-4, 136 Joystick, 206

INITIALIZE-S, 159
INITIALIZE-6, 196 Keyboard, 212
INITIALIZE-7, 224 Koch curve, 426
INITIALIZE-8, 301
INITIALIZE-9, 340
Lagrange interpolation, 403
INITIALIZE-9A, 318
Lambert’s law, 346
INITIALIZE-10, 362
Landscape mode, 118
Ink-jet printer, 37
Laser printer, 36-37
Inking, 232
Left-handed coordinates, 245
Input devices, 206-211, 239
Length:
Input/output statement, 451
line segment, 6
INQUIRE-BACK-CLIPPING, 308
vector, 10
INQUIRE-CHARSIZE, 64
Light button, 239
INQUIRE-CHARSPACE, 64
Light pen, 208
INQUIRE-CHARUP, 64
INQUIRE-DEPTH-CLIPPING-PLANES,
LINE-ABS-2, 44
308 LINE-ABS-3, 250
INQUIRE-FILL-STYLE, 99 LINE-REL-2, 45
INQUIRE-FILLED, 99 LINE-REL-3, 250
Lines:
INQUIRE-FRONT-CLIPPING, 308
INQUIRE-IMAGE-TRANSFORMATION, antialiasing, 20-21, 26
169 equations for, 3-6
INQUIRE-LIGHT, 395 general form of, 4
INQUIRE-LINESTYLE, 64 intersections of, 4—5
INQUIRE-OBJECT-SHADE, 395 parametric form, 6, 246
INQUIRE-PARALLEL, 308 perpendicular, 7-8
INQUIRE-PERSPECTIVE, 308 position relative to, 197
INQUIRE-ROTATE, 143 scan, 79
INQUIRE-SCALE, 143 segments, 5—7
INQUIRE-SEGMENT-OPEN, 169 slope-intercept form, 4
INQUIRE-SHADING, 395 style, 55
INQUIRE-TRANSLATE, 143 thick, 21
INQUIRE-VIEW-DISTANCE, 308 three-dimensional, 246
INQUIRE-VIEW-PLANE-NORMAL, 308 vector form, 11
INQUIRE-VIEW-REFERENCE-POINT, 308 Linked list, 161
INQUIRE-VIEW-UP, 308 Liquid crystal display, 35
INQUIRE-VISIBILITY, 169 List of available space, 161
Inside test, 74-76, 96 LOAD-POLYGON, 84, 130
INTENSITY, 360 Locator, 206, 212
Interactive computer graphics, 1 simulating with a pick, 225
Interactive techniques, 230-240 Luminosity, 376
Intercept, 4
Interior edge, 326 Mach banding, 362
Interpolation, 400 MAKE-CURVE, 405
INTERPRET, 47, 53, 56, 78, 129 MAKE-PARALLEL-
Interrupt, 209 TRANSFORMATION, 284
Intersection of lines, 4—5 MAKE-PERSPECTIVE-
Isometric projection, 277 TRANSFORMATION, 284
 INDEX 463

MAKE-PICTURE-CURRENT, 48, 129, Operand, 42

160 Operation code, 42
MAKE-SHADE, 358 Operators, 448
MAKE-VIEW-PLANE- Ordered dither, 374, 391
TRANSFORMATION, 283 Outcodes, 182
MAKE-Z-CLIP-PLANES, 288
Marker, 65 Paging, 162
MARKER-ABS-2, 65 Painter’s algorithm, 321, 341
MARKER-REL-2, 65 Painting, 233
Matrix, 108 Parallel projection, 261-263
determinant, 124—125 PARALLEL-TRANSFORM, 264
homogeneous coordinate, 117, 120 Parametric form, 6, 246, 400
identity, 109 Patches, 417
inverse, 122-125 Patterns, 91
multiplication, 108-109, 119 aliasing, 136, 140
parallel projection, 263 registration, 91
perspective projection, 265-266 replication, 91, 93
post-multiplication, 119 transformation, 131-136, 140
pre-multiplication, 119 Peano curve, 425
rotation, 113-115, 117, 119-120 Perpendicular lines, 7-8
scaling, 109-112, 117 Perspective projection, 264-267, 279
translation, 117 PERSPECTIVE-TRANSFORM, 267
Menu, 239 PHIGS (programmer’s hierarchical
Metafile, 36, 41 interactive graphics standard), 2
Midpoint, 7 Photopigments, 376
Minimax test, 322 Pick, 212, 240
Mouse, 207, 240 simulating with a locator, 225
MOVE-ABS-2, 44 PICK-SEARCH, 229
MOVE-ABS-3, 250 Pixel, 11
MOVE-REL-2, 45 Plane, 246-247
MOVE-REL-3, 250 Plasma panel, 35
Multiplication, matrix, 108-109, 119 Plotter, 34
MULTIPLY-IN-ROTATION, 127 Point plotting, 231
MULTIPLY-IN-SCALE, 126 Point-source illumination, 348
MULTIPLY-IN-TRANSLATION, 126 Polling loop, 209
POLY-INSERT, 85
NEW-FRAME, 48 Polygon, 70
NEW-TRANSFORM-3, 254 clipping, 190-195
NEW-VIEW-2, 179 concave, 71
NEW-VIEW-3, 300, 362 convex, 71
NEXT-SECTION, 406 edges, 71
Normal to a plane, 247 filling, 79-93, 96
Normalized device coordinates, 41 inside test, 74-76, 96
opcodes, 71
Object coordinates, 269 representations, 96
Object space, 173 vertex, 71
Oblique projection, 278 POLYGON-ABS-2, 73
One-point perspective projection, 279 POLYGON-ABS-3, 251
Opcode, 42 POLYGON-REL-2, 74
464 inpex

POLYGON-REL-3, 251 Refraction, 365

POLYLINE-ABS-2, 64 Relative line command, 38
POLYLINE-REL-2, 64 Relative move command, 38
POLYMARKER-ABS-2, 65 RENAME-SEGMENT, 154, 223
POLYMARKER-REL-2, 65 Resolution, 11,34
Portrait mode, 118 RGB color model, 379
Post-multiplication, 119 RGB-TO-HSV, 385
Pre-multiplication, 119 Right-handed coordinates, 245
Primitives, 37-39, 49, 55, 60 ROTATE, 128, 158
Procedure, 450 ROTATE-X-3, 255
Programmer’s hierarchical interactive ROTATE-Y-3, 256
graphics standard (PHIGS), 2 ROTATE-Z-3, 256
Projection, 261, 304 Rotation, 113-115, 253-254, 304
axonometric, 276 about an arbitrary axis, 256-261
cabinet, 278 about an arbitrary point, 119-120
cavalier, 278 Rubber band lines, 235
center of, 264
dimetric, 278 Sampled device, 211
isometric, 277 Saturation, 378, 383, 385
oblique, 278 SAVE-CLIPPED-POINT, 190, 296
parallel, 261-263 SAVE-POLY-FOR-HSC, 330
perspective, 264-267 SAVE-POLYGONS-IN-ORDER, 339
one-point, 279 SCALE, 128, 158
three-point, 279 Scales, 238-239
two-point, 279 Scaling, 109-112, 252
trimetric, 278 Scan-line algorithms, 79
Pseudo display file, 36 Segments:
PUT-IN-D, 338 display file, 146, 166
PUT-IN-E, 339 closing, 150
PUT-IN-SM, 408 creation, 149
PUT-IN-T, 191, 297 deletion, 150—153
PUT-POINT, 43, 159 name, 147
Pythagorean theorem, 6 renaming, 153-154
table for, 147-149
Quantitative invisibility, 326 unnamed, 148
Queue, 215 visibility, 154-155
Ine, 5=7; 21
Raster display, 34 Selector, 206
Raster techniques, 163-165 SEND-TO-DF, 339% ~*
RasterOp, 164, 166 SET-B-SPLINE, 413 ~
Ray tracing, 370-371, 391, 438 SET-BACK-PLANE-CLIPPING, 288
READ-LOCATOR, 221 SET-CHARSPACE, 52
READ-VALUATOR, 222 SET-CHARUP, 52
Reflection, 304, 367 SET-DETECTABILITY, 222
angle of, 350 SET-FILL, 77
coefficient of, 346 SET-FILL-STYLE, 77
specular, 349 SET-FRONT-PLANE-CLIPPING, 288
Reflection transformations, 121, 304 SET-HIDDEN-LINE-REMOVAL, 316
Reflectivity, 346 SET-IMAGE-TRANSFORMATION, 156
 INDEX 465

SET-IMAGE-TRANSLATION, 156 Tablet, 207, 239

SET-LIGHT, 361 TERMINATE, 26, 454
SET-LINESTYLE, 56 Text, 49-55
SET-MARKER-SYMBOL, 65 TEXT, 53
SET-OBJECT-SHADE, 360 Texture, 391, 438
SET-PARALLEL, 276 Thermal printer, 37
SET-PATTERN-REPRESENTATION, 92 Three-point perspective projection, 279
SET-PERSPECTIVE, 276 Thumbwheels, 206
SET-SHADING, 355 Topological dimension, 424
SET-SMOOTH, 404 Touch panel, 240
SET-TRANSFORM-PATTERN, 134 Trackball, 207
SET-UP-SPLIT-POLY, 335 Tracking cross, 225
SET-VIEW-DEPTH, 287 Transformation, 107, 140, 304
SET-VIEW-DISTANCE, 274 coordinate, 118, 133-134
SET-VIEW-PLANE-NORMAL, 274 homogeneous, 117, 120
SET-VIEW-REFERENCE-POINT, 272 image, 155-157
SET-VIEW-UP, 274 inverse, 122—125
SET-VIEWPORT, 178 parallel projection, 263
SET-VISIBILITY, 155 patterns, 131-136, 140
SET-WINDOW, 178 perspective projection, 265-266
SHADE-CHECK, 356 reflection, 121, 304
Shading function, 354, 390 rotation, 113-115, 253-254, 304
Shading models, 391 scaling, 109-112, 252
Shadow volumes, 368 shear, 121
Shadows, 368-369, 391 three-dimensional, 251-254
Shear, 121 translation, 116-118, 252
SIGNOF, 338 viewing, 175-178, 298-301
Sine, 112-113 TRANSLATE, 128, 157
Slope, 4 TRANSLATE-3, 255
Slope-intercept form, 4 Translation, 116-118, 252
SMOOTH-POLY-ABS-3, 409 Transparency, 365-367, 391
Snell’s law, 365 Transparency coefficient, 365
Solid modeling, 304 Trapezoid, 71, 96
SORT-POLYGONS, 331 Trimetric projection, 278
Specular reflection, 349 Two-point perspective projection, 279
START-B-SPLINE, 414
START-CURVE, 406 UNITY, 355
Statement, 449 Unnamed segment, 148
assignment, 449 UPDATE-X-VALUES, 90
compound, 449
control, 449 Valuator, 212, 240
input/output, 451 Variables, 447
Stereo views, 304, 309 VECGEN, 15
Stroke, 233 Vector, 10-11
Stroke characters, 22 Vector cross product, 313
Stylus, 207 Vector dot product, 247, 315
Surface patch, 417 Vector generation, 11-20, 26
Sutherland-Hodgman algorithm, 184, 200 Vector normal, 247, 313-314
Synthetic camera, 268 Vector refresh display, 35-36
466 INDEX

Vertices, 71 Warnock’s algorithm, 323, 341

View distance, 268 Wavelength, 376
Viewing parameters, 268-272, 304 Winding-number inside test, 75
VIEWING-TRANSFORM, 181 Window, 173
Viewing transformation, 175-178, Windowing, 172
298-301 multiple, 198
View plane coordinates, 268, 279-282 Wire-frame drawings, 311
View plane normal, 268
VIEW-PLANE-TRANSFORM,, 299 XSORT, 88
View reference point, 268
View-up direction, 270
View volume, 286 YIQ color model, 382
Viewport, 173
Visibility, 146 Z Buffer, 319, 341, 369, 391
i=
ed me

= : 2

vo

*
poe:Ara— oP
ee = ?
 QW
OTHER McGRAW-HILL
INTERNATIONAL EDITIONS
IN RELATED FIELDS

Chapra: INTRODUCTION TO COMPUTING FOR ENGINEERS

Davis: INTRODUCTION TO COMPUTERS, 3/e
Gore-Stubbe: COMPUTERS AND INFORMATION SYSTEMS, 2/e
Saret: DATA PROCESSING LOGIC, 2/e
Thomas: ADVANCED PROGRAMMER'S GUIDE TO UNIX SYSTEM V
Voisinet. COMPUTER-AIDED DRAFTING AND DESIGN: Concepts and
Applications
Voisinet: INTRODUCTION TO CAD, 2/e
Leventhal: 68000 ASSEMBLY LANGUAGE PROGRAMMING, 2/e
Lipschutz: ESSENTIAL COMPUTER MATHEMATICS [SCHAUM]
Newman: DESIGNING INTEGRATED SYSTEMS FOR THE OFFICE
ENVIRONMENT
Newman-Sproull: PRINCIPLES OF INTERACTIVE COMPUTER
GRAPHIGS, 2/e
Orilias COMPUTERS AND INFORMATION: An Introduction, 3/e
Philippakis-Kazmier: ADVANCED COBOL, 2/e
Plastock: COMPUTER GRAPHICS [SCHAUM]
Pressman: SOFTWARE ENGINEERING, 2/e
Rogers: PROCEDURAL ELEMENTS FOR COMPUTER GRAPHICS
Sanders: COMPUTERS IN BUSINESS, 4/e

ISBN 0-07-100472 -i=

gat

Mk Irs Say

liz
Sv

KO
1