Human Visual Perception

The Human Eye

• Diameter: 20 mm

• 3 membranes enclose the eye

– Cornea & sclera
– Choroid
– Retina
 The Choroid
• The choroid contains blood vessels for eye
nutrition and is heavily pigmented to reduce
extraneous light entrance and backscatter.

• It is divided into the ciliary body and the iris

diaphragm, which controls the amount of
light that enters the pupil (2 mm ~ 8 mm).
 The Lens
• The lens is made up of fibrous cells and is
suspended by fibers that attach it to the
ciliary body.

• It is slightly yellow and absorbs approx. 8%

of the visible light spectrum.
 The Retina
• The retina lines the entire posterior portion.

• Discrete light receptors are distributed over

the surface of the retina:

– cones (6-7 million per eye) and

– rods (75-150 million per eye)
 Cones
• Cones are located in the fovea and are
sensitive to color.

• Each one is connected to its own nerve end.

• Cone vision is called photopic (or bright-light

vision).
 Rods
• Rods are giving a general, overall picture of
the field of view and are not involved in
color vision.

• Several rods are connected to a single nerve

and are sensitive to low levels of
illumination (scotopic or dim-light vision).
 Receptor Distribution
• The distribution of receptors is radially
symmetric about the fovea.

• Cones are most dense in the center of the

fovea while rods increase in density from the
center out to approximately 20% off axis and
then decrease.
Cones & Rods
 The Fovea
• The fovea is circular (1.5 mm in diameter)
but can be assumed to be a square sensor
array (1.5 mm x 1.5 mm).

• The density of cones: 150,000 elements/mm2

~ 337,000 for the fovea.

• A CCD imaging chip of medium resolution

needs 5 mm x 5 mm for this number of
elements
 Image Formation in the Eye
• The eye lens (if compared to an optical lens)
is flexible.

• It gets controlled by the fibers of the ciliary

body and to focus on distant objects it gets
flatter (and vice versa).
 Image Formation in the Eye
• Distance between the center of the lens and
the retina (focal length):
– varies from 17 mm to 14 mm (refractive power of
lens goes from minimum to maximum).

• Objects farther than 3 m use minimum

refractive lens powers (and vice versa).
 Image Formation in the Eye
• Example:
– Calculation of retinal image of an object

15 x
=
100 17

x = 2.55mm
 Image Formation in the Eye
• Perception takes place by the relative
excitation of light receptors.

• These receptors transform radiant energy

into electrical impulses that are ultimately
decoded by the brain.
 Brightness Adaptation &
Discrimination
• Range of light intensity levels to which HVS
(human visual system) can adapt: on the
order of 1010.

• Subjective brightness (i.e. intensity as

perceived by the HVS) is a logarithmic
function of the light intensity incident on the
eye.
 Brightness Adaptation &
Discrimination
• The HVS cannot operate over such a range
simultaneously.

• For any given set of conditions, the current

sensitivity level of HVS is called the
brightness adaptation level.
 Brightness Adaptation &
Discrimination
• The eye also discriminates between changes
in brightness at any specific adaptation
level.
ΔI c
→ Weber ratio
I
Where: ΔIc: the increment of illumination
discriminable 50% of the time and
I : background illumination
 Brightness Adaptation &
Discrimination
• Small values of Weber ratio mean good
brightness discrimination (and vice versa).

• At low levels of illumination brightness

discrimination is poor (rods) and it improves
significantly as background illumination
increases (cones).
 Brightness Adaptation &
Discrimination
• The typical observer can discern one to two
dozen different intensity changes

– i.e. the number of different intensities a person

can see at any one point in a monochrome image
 Brightness Adaptation &
Discrimination
• Overall intensity discrimination is broad due
to different set of incremental changes to be
detected at each new adaptation level.

• Perceived brightness is not a simple function

of intensity
– Scalloped effect, Mach band pattern
– Simultaneous contrast
Perceived Brightness
Simultaneous Contrast
Illusions
 A Simple Image Model
• Image: a 2-D light-intensity function f(x,y)

• The value of f at (x,y) Æ the intensity

(brightness) of the image at that point

• 0 < f(x,y) < ∞

Digital Image Acquisition
 A Simple Image Model
• Nature of f(x,y):

– The amount of source light incident on the scene

being viewed

– The amount of light reflected by the objects in

the scene
 A Simple Image Model
• Illumination & reflectance components:

– Illumination: i(x,y)
– Reflectance: r(x,y)

– f(x,y) = i(x,y) ⋅ r(x,y)

– 0 < i(x,y) < ∞

and 0 < r(x,y) < 1
(from total absorption to total reflectance)
 A Simple Image Model
• Sample values of r(x,y):
– 0.01: black velvet
– 0.93: snow

• Sample values of i(x,y):

– 9000 foot-candles: sunny day
– 1000 foot-candles: cloudy day
– 0.01 foot-candles: full moon
 A Simple Image Model
• Intensity of a monochrome image f at (xo,yo):
gray level l of the image at that point
l=f(xo, yo)

• Lmin ≤ l ≤ Lmax
– Where Lmin: positive
Lmax: finite
 A Simple Image Model
• In practice:
– Lmin = imin rmin and
– Lmax = imax rmax

• E.g. for indoor image processing:

– Lmin ≈ 10 Lmax ≈ 1000

• [Lmin, Lmax] : gray scale

– Often shifted to [0,L-1] Æ l=0: black
l=L-1: white
 Sampling & Quantization
• The spatial and amplitude digitization of
f(x,y) is called:

– image sampling when it refers to spatial

coordinates (x,y) and

– gray-level quantization when it refers to the

amplitude.
Digital Image
Sampling and Quantization
A Digital Image
 Sampling & Quantization
⎡ f (0,0) f (0,1) ... f (0, M − 1) ⎤
⎢ f (1,0) ... ... f (1, M − 1) ⎥⎥
f ( x, y ) = ⎢
⎢ ... ... ... ... ⎥
⎢ ⎥
⎣ f ( N − 1,0) f ( N − 1,1) ... f ( N − 1, M − 1)⎦

Digital Image Image Elements

(Pixels)
 Sampling & Quantization

• Important terms for future discussion:

– Z: set of real integers

– R: set of real numbers

 Sampling & Quantization

• Sampling: partitioning xy plane into a grid

– the coordinate of the center of each grid is a pair

of elements from the Cartesian product Z x Z (Z2)

• Z2 is the set of all ordered pairs of elements

(a,b) with a and b being integers from Z.
 Sampling & Quantization
• f(x,y) is a digital image if:

– (x,y) are integers from Z2 and

– f is a function that assigns a gray-level value
(from R) to each distinct pair of coordinates (x,y)
[quantization]

• Gray levels are usually integers

– then Z replaces R
 Sampling & Quantization
• The digitization process requires decisions
about:

– values for N,M (where N x M: the image array)

and

– the number of discrete gray levels allowed for

each pixel.
 Sampling & Quantization
• Usually, in DIP these quantities are integer
powers of two:
N=2n M=2m and G=2k

number of gray levels

• Another assumption is that the discrete

levels are equally spaced between 0 and L-1
in the gray scale.
Examples
Examples
Examples
Examples
 Sampling & Quantization
• If b is the number of bits required to store a
digitized image then:

– b = N x M x k (if M=N, then b=N2k)

Storage
 Sampling & Quantization
• How many samples and gray levels are
required for a good approximation?

– Resolution (the degree of discernible detail) of

an image depends on sample number and gray
level number.
– i.e. the more these parameters are increased,
the closer the digitized array approximates the
original image.
 Sampling & Quantization
• How many samples and gray levels are
required for a good approximation? (cont.)

– But: storage & processing requirements increase

rapidly as a function of N, M, and k
 Sampling & Quantization
• Different versions (images) of the same
object can be generated through:

– Varying N, M numbers
– Varying k (number of bits)
– Varying both
 Sampling & Quantization
• Isopreference curves (in the Nm plane)

– Each point: image having values of N and k equal

to the coordinates of this point

– Points lying on an isopreference curve correspond

to images of equal subjective quality.
Examples
Isopreference Curves
 Sampling & Quantization

• Conclusions:
– Quality of images increases as N & k increase
– Sometimes, for fixed N, the quality improved by
decreasing k (increased contrast)
– For images with large amounts of detail, few gray
levels are needed
 Nonuniform
Sampling & Quantization
• An adaptive sampling scheme can improve the
appearance of an image, where the sampling would
consider the characteristics of the image.

– i.e. fine sampling in the neighborhood of sharp gray-level

transitions (e.g. boundaries)
– Coarse sampling in relatively smooth regions

• Considerations: boundary detection, detail content

 Nonuniform
Sampling & Quantization
• Similarly: nonuniform quantization process

• In this case:
– few gray levels in the neighborhood of
boundaries
– more in regions of smooth gray-level variations
(reducing thus false contours)
 Some Basic Relationships
Between Pixels
• Definitions:

– f(x,y): digital image

– Pixels: q, p
– Subset of pixels of f(x,y): S
 Neighbors of a Pixel
• A pixel p at (x,y) has 2 horizontal and 2
vertical neighbors:

– (x+1,y), (x-1,y), (x,y+1), (x,y-1)

– This set of pixels is called the 4-neighbors

of p: N4(p)
 Neighbors of a Pixel
• The 4 diagonal neighbors of p are: (ND(p))

– (x+1,y+1), (x+1,y-1), (x-1,y+1), (x-1,y-1)

• N4(p) + ND(p) Æ N8(p): the 8-neighbors of p

 Connectivity
• Connectivity between pixels is important:

– Because it is used in establishing boundaries of

objects and components of regions in an image
 Connectivity
• Two pixels are connected if:

– They are neighbors (i.e. adjacent in some sense -

- e.g. N4(p), N8(p), …)
– Their gray levels satisfy a specified criterion of
similarity (e.g. equality, …)

• V is the set of gray-level values used to

define adjacency (e.g. V={1} for adjacency
of pixels of value 1)
 Adjacency
• We consider three types of adjacency:

– 4-adjacency: two pixels p and q with values from

V are 4-adjacent if q is in the set N4(p)

– 8-adjacency : p & q are 8- adjacent if q is in the

set N8(p)
 Adjacency
• The third type of adjacency:

– m-adjacency: p & q with values from V are m-

adjacent if

• q is in N4(p) or
• q is in ND(p) and the set N4(p)∩N4(q) has no pixels with
values from V
 Adjacency
• Mixed adjacency is a modification of 8-
adjacency and is used to eliminate the
multiple path connections that often arise
when 8-adjacency is used.

0 1 1 0 1 1 0 1 1
0 1 0 0 1 0 0 1 0
0 0 1 0 0 1 0 0 1
 Adjacency

• Two image subsets S1 and S2 are adjacent if some

pixel in S1 is adjacent to some pixel in S2.
 Path
• A path (curve) from pixel p with coordinates
(x,y) to pixel q with coordinates (s,t) is a
sequence of distinct pixels:

– (x0,y0), (x1,y1), …, (xn,yn)

– where (x0,y0)=(x,y), (xn,yn)=(s,t), and

(xi,yi) is adjacent to (xi-1,yi-1), for
1≤i ≤n ; n is the length of the path.
• If (xo, yo) = (xn, yn): a closed path
 Paths
• 4-, 8-, m-paths can be defined depending on
the type of adjacency specified.

• If p,q ∈ S, then q is connected to p in S if

there is a path from p to q consisting
entirely of pixels in S.
 Connectivity
• For any pixel p in S, the set of pixels in S
that are connected to p is a connected
component of S.

• If S has only one connected component then

S is called a connected set.
 Boundary
• R a subset of pixels: R is a region if R is a
connected set.
• Its boundary (border, contour) is the set of
pixels in R that have at least one neighbor
not in R
• Edge can be the region boundary (in binary
images)
 Distance Measures
• For pixels p,q,z with coordinates (x,y), (s,t),
(u,v), D is a distance function or metric if:

– D(p,q) ≥ 0 (D(p,q)=0 iff p=q)

– D(p,q) = D(q,p) and
– D(p,z) ≤ D(p,q) + D(q,z)
 Distance Measures
• Euclidean distance:

– De(p,q) = [(x-s)2 + (y-t)2]1/2

– Points (pixels) having a distance less than or

equal to r from (x,y) are contained in a disk of
radius r centered at (x,y).
 Distance Measures
• D4 distance (city-block distance):

– D4(p,q) = |x-s| + |y-t|

– forms a diamond centered at (x,y)
– e.g. pixels with D4≤2 from p
2
2 1 2
2 1 0 1 2 D4 = 1 are the 4-neighbors of p
2 1 2
2
 Distance Measures
• D8 distance (chessboard distance):

– D8(p,q) = max(|x-s|,|y-t|)
– Forms a square centered at p
– e.g. pixels with D8≤2 from p
2 2 2 2 2
2 1 1 1 2
2 1 0 1 2 D8 = 1 are the 8-neighbors of p
2 1 1 1 2
2 2 2 2 2
 Distance Measures
• D4 and D8 distances between p and q are
independent of any paths that exist between
the points because these distances involve
only the coordinates of the points
(regardless of whether a connected path
exists between them).
 Distance Measures
• However, for m-connectivity the value of the
distance (length of path) between two pixels
depends on the values of the pixels along the
path and those of their neighbors.
 Distance Measures
• e.g. assume p, p2, p4 = 1
p1, p3 = can have either 0 or 1

If only connectivity of pixels valued 1 is

p3 p4 allowed, and p1 and p3 are 0, the m-
distance between p and p4 is 2.
p1 p2
If either p1 or p3 is 1, the distance is 3.
p
If both p1 and p3 are 1, the distance is 4
(pp1p2p3p4)