Unit 3 (Slides) PDF

DIGITAL IMAGE PROCESSING - 2
Lecture 24: Image Segmentation
Dr. Shruthi M.L.J.

Department of Electronics &
Communication Engineering
1
Unit 3: Image Segmentation
Dr. Shruthi M.L.J.

Department of Electronics & Communication Engineering
2
Last Session
• Grayscale Morphology Operations

• Erosion
• Dilation
• Opening
• Closing
3
Image Segmentation
• Image segmentation is an essential preliminary step in most

automatic pictorial pattern recognition and scene analysis
algorithms.
• Segmentation is to subdivide an image into constituent regions

or objects.
• Image segmentation algorithms are generally based on one of

the two basic properties of intensity value: similarity and
discontinuity.
4
Image Segmentation
• Segmentation techniques are either region-based or edge-

based
• Region-based techniques rely on common patterns in

intensity values within a cluster of neighboring pixels.
• The cluster is referred to as the region, and the goal of
the segmentation algorithm is to group regions
according to their anatomical or functional roles
• Edge-based techniques rely on discontinuities in image

values between distinct regions, and the goal of the
segmentation algorithm is to accurately demarcate the
boundary separating these regions
5
Image Segmentation
To segment the image,

we assign one level
(say, white) to the
pixels on or inside the
boundary, and another
Image of a constant Boundary based on level (e.g., black) to all
intensity discontinuities. Result of segmentation.
intensity region. points exterior to the
boundary.
Result of intensity discontinuity Result of segmentation 6

Image of a texture region computations based on region properties.
Image Segmentation
• Two types
• Local Segmentation
• Deals with sub-images which are small windows of the whole
image
• Number of pixels available is much lower than global
segmentation
• Global Segmentation
• Concerned with segmenting the whole image
• Deals with segments consisting of a relatively large number of
pixels hence estimated parameter values are more robust
7
Image Segmentation Techniques

• Three Approaches
• Points
• Lines
• Edges
8
Edge-Based Segmentation
• Deals with detection of sharp, local changes in intensity
• Three basic types of gray-level discontinuities in a digital image
are:
• Isolated Points
• Foreground (background) pixel surrounded by
background (Foreground) pixels
• Lines
• Thin edge segment in which intensity of the background on
either side of the line is either much higher or much lower
than intensity of the line pixels
• Edges
• Edges: Set of connected edge pixels
• Edge Pixels are pixels at which intensity of image changes
abruptly
9
Isolated Point, Line and Edge
10
Basic Concept
• The most common way to look for discontinuities is to run a

mask through the image.
• The response of the filter at the center point of the kernel is
w1 w2 w3
w4 w5 W6
w7 w8 w9
Point Detection Mask
11
Next Session
• Segmentation
• Edge Approach
12
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
13
Lecture 25: Image Segmentation
Dr. Shruthi M.L.J.

1
Unit 3: Image Segmentation
Dr. Shruthi M.L.J.

2
Last Session
• Images segmentation
• Types of segmentation
• Local
• Global
• Types of segmentation techniques
• Edge approach
• Boundary approach
• Region approach
3
Today’s Session
• Types of segmentation techniques

• Edge approach
• Boundary approach
• Region approach
4
Image Segmentation
• Two types
• Local Segmentation
• Deals with sub-images which are small windows of the whole image
• Number of pixels available is much lower than global segmentation
• Global Segmentation
• Concerned with segmenting the whole image
• Deals with segments consisting of a relatively large number of pixels
5
• Segmentation techniques are either region-based or edge-

based
• Region-based techniques rely on common patterns in

intensity values within a cluster of neighboring pixels.
• The cluster is referred to as the region, and the goal of
the segmentation algorithm is to group regions
according to their anatomical or functional roles
• Edge-based techniques rely on discontinuities in image

values between distinct regions, and the goal of the
segmentation algorithm is to accurately demarcate the
boundary separating these regions
6
Isolated Point, Line and Edge
7
Derivatives of digital functions/images
• Derivatives of a digital function are defined in terms of finite

differences.
• Approximation used for First derivative:
• must be zero in areas of constant intensity
• must be nonzero at the onset of an intensity step or ramp; and
• must be nonzero at points along an intensity ramp
• Approximation used for second derivatives
• must be zero in areas of constant intensity
• must be nonzero at the onset and end of an intensity step or ramp
• must be zero along intensity ramps
• Because we are dealing with digital quantities whose values
are finite, the maximum possible intensity change is also finite,
and the shortest distance over which a change can occur is
between adjacent pixels 8
First Order Derivatives of digital functions/images
• Using Taylor series expansion we obtain approximations to the
first-order derivative
• First-order derivative at an arbitrary point x of a one-
dimensional function f (x) is obtained by expanding the
function into a Taylor series about x
• Where is the separation between samples of f.

• Here this separation is measured in pixel units.
• Thus, following the convention in the book, = -1 for the
sample preceding x and = 1 for the sample following x
9
First Order Derivatives of digital functions/images
• First-order derivative at an arbitrary point x of a one-

dimensional function f (x) is given by
for
for
10
Derivatives using approximations to Taylor Series
• We compute intensity differences using just a few terms of the

Taylor series.
• For first-order derivatives we use only the linear terms, and we
can form differences in one of three ways:
• Forward Difference(keeping only linear terms):
• Backward Difference:
• Central Difference(obtained from Taylor series by subtracting

first order derivatives(backward from forward difference):
11
Derivatives using approximations to Taylor Series
• In general, the more terms we use from the Taylor series to

represent a derivative, the more accurate the approximation
will be
• It turns out that central differences have a lower error for the
same number of points
• For this reason, derivatives are usually expressed as central differences.
• The second order derivative based on a central difference,
∂2 f (x)/∂x2 is obtained by adding backward difference and
forward difference equations(neglecting higher order terms)
12
Derivatives in digital image
• Determine the first and second order derivatives of the following image
Horizontal intensity profile

that includes the isolated point
indicated by the arrow
Subsampled profile 13
Derivatives in digital image
• The transition in the ramp spans four pixels, the noise point is a single pixel,
• the line is three pixels thick, and the transition of the step edge takes place
between adjacent pixels.
• The number of intensity levels was limited to eight for simplicity. 14
Properties of first and second Derivatives in digital image
• Initially, the first-order derivative is nonzero at the onset and along

the entire intensity ramp, while the second-order derivative is
nonzero only at the onset and end of the ramp.
• Because the edges of digital images resemble this type of transition,
we conclude that first-order derivatives produce “thick” edges, and
second-order derivatives much thinner ones.
15
• Next we encounter the isolated noise point.

• Here, the magnitude of the response at the point is much stronger for the
second than for the first-order derivative.
• A second-order derivative is much more aggressive than a first-order
derivative in enhancing sharp changes.
• Thus, we can expect second-order derivatives to enhance fine detail
(including noise) much more than first-order derivatives. 16
• The line in this example is rather thin, so it too is fine detail, and we see
again that the second derivative has a larger magnitude.
• At both the ramp and step edges the second derivative has opposite signs
(negative to positive or positive to negative) as it transitions into and out
of an edge. This “double-edge” effect is an important characteristic that
can be used to locate edges
• As we move into the edge, the sign of the second derivative is used also to
determine whether an edge is a transition from light to dark (negative
17
second derivative), or from dark to light (positive second derivative)
Summary of first and second Derivatives in digital image
• First-order derivatives generally produce thicker edges
• Second-order derivatives have a stronger response to

fine detail, such as thin lines, isolated points, and noise.
• Second-order derivatives produce a double-edge

response at ramp and step transitions in intensity
• The sign of the second derivative can be used to

determine whether a transition into an edge is from
light to dark or dark to light.
18
Basic Concept
• The approach of choice for computing first and second

derivatives at every pixel location in an image is to use spatial
convolution.
• For the 3 × 3 filter kernel shown, the procedure is to compute
the sum of products of the kernel coefficients with the
intensity values in the region encompassed by the kernel
• That is, the response of the filter at the center point of the
kernel is
where zk is the intensity of the pixel whose spatial location

corresponds to the location of the kth kernel coefficient. 19

20
are:
• Isolated Points
• Lines
• Edges
abruptly
21
Detection of Isolated Point
• Point Detection:
• Point detection should be based on the second derivative using
the Laplacian:
where the partial derivatives are computed using the second-

order finite differences.
The Laplacian is then
• This expression can be implemented using the Laplacian kernel

1 1 1
1 8 1
1 1 1 22
• A point has been detected at a location (x, y) on which the

kernel is centered if the absolute value of the response of the
filter at that point exceeds a specified threshold.
• Such points are labeled 1 and all others are labeled 0 in the
output image, thus producing a binary image
• Use the expression
where g(x, y) is the output image, T is a nonnegative threshold,

and Z is given as indicated
23
• This formulation simply measures the weighted differences

between a pixel and its 8-neighbors
• Isolated point will be quite different from its surroundings

hence easily detectable by this mask
• Note that, as usual for a derivative kernel, the coefficients sum

to zero, indicating that the filter response will be zero in areas of
constant intensity
24
Next Session
• Image Segmentation Cont..

• Point detection
• Line detection
25
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
26
U3 Lec 26-27: Image Segmentation
Dr. Shruthi M.L.J.

1
Dr. Shruthi M.L.J.

2

3
are:
• Isolated Points
• Lines
• Edges
abruptly
4
• Point Detection:
• Point detection is based on the second derivative using the
Laplacian:
where the partial derivatives are computed using the second-order

finite differences.
The Laplacian is then
• This expression can be implemented using the Laplacian kernel

1 1 1
1 8 1
1 1 1 5
• A point has been detected at a location (x, y) on which the

kernel is centered if the absolute value of the response of the
filter at that point exceeds a specified threshold.
• Such points are labeled 1 and all others are labeled 0 in the
output image, thus producing a binary image
• Use the expression
where g(x, y) is the output image, T is a nonnegative threshold,

and Z is given as indicated
6
• This formulation simply measures the weighted differences

between a pixel and its 8-neighbors
• Isolated point will be quite different from its surroundings

hence easily detectable by this mask
• Note that, as usual for a derivative kernel, the coefficients sum

to zero, indicating that the filter response will be zero in areas of
constant intensity
7
Point Detection Example
1 1 1
1 8 1
1 1 1
Laplacian kernel
Result of Result of using used
X-ray image of a turbine
convolving the thresholding where a for point detection
blade with a porosity
kernel with the single point is detected
manifested by a single
image (shown enlarged at the
black pixel
tip of the arrow)
8

• Three basic types of gray-level discontinuities in a digital image are:
Isolated Points
• Foreground (background) pixel surrounded by background
(Foreground) pixels
• Lines
• Thin edge segment in which intensity of the background on either
side of the line is either much higher or much lower than intensity
of the line pixels
• Edges
• Edge Pixels are pixels at which intensity of image changes abruptly
9
Coordinate Convention in Digital Image
Coordinate convention used to represent

digital images. Because coordinate values
are integers, there is a one-to-one
correspondence between x and y and the
rows (r) and columns (c) of a matrix. 10
Line Detection
• Consider kernels as shown below:

Detection angles are with respect to the axis
system where positive
angles measured counter clockwise with
respect to the (vertical) x-axis.
Horizontal +450 Vertical - 450
Line detection kernels.
• First mask would respond more strongly to lines oriented

horizontally
• Maximum response would result when line passes through
middle row
• Second would respond best to line oriented at +450 and so on
11
Line Detection Procedure
• Let Z1, Z2, Z3 and Z4 be the responses to masks from left to right
• Let the masks be run individually through
with these the
fourimage
• If at a certain point Z k > Z j for all j ¹ k
that point is said toinbethe tionassociated
direc-
more likely of kernel with a line in the
direction of the kernel k
• For example, if at a point in the image, Z 1 > Z j for
j = 2,3,4, that point is said to be more likely associated with a
horizontal line
12
Line Detection Procedure
• If we are interested in detecting all the lines in an image in the

direction defined by a given kernel, we simply run the kernel
through the image and threshold the absolute value of the
result
• The nonzero points remaining after thresholding are the

strongest responses which, for lines one pixel thick, correspond
closest to the direction defined by the kernel
13
Line Detection Example
• Determine lines that are

one pixel thick and
oriented towards + 450
• To get one pixel thick line
we use threshold equal
to (maximum value-1)
(254)
14
• we use the following kernel
the shades darker than the gray

background correspond to negative
values
The result of filtering the image with this kernel. 15

The result of filtering There are two principal segments in the image oriented in
the image with this kernel. the +45 direction, one in the top left and one at the
bottom right.
The straight line segment in third figure is brighter than the
segment second figure because the line segment in the
bottom right of given figure is one pixel thick, while the
one at the top left is not.
The kernel is “tuned” to detect one-pixel-thick lines in the
+45 direction, so we expect its response to be stronger 16
when such lines are detected
• Next we apply the threshold for stronger response by letting T=254
The result of filtering

The image with all negative All points (in white) whose
the image with this kernel. values satisfied the condition
values set to zero.
g > T,
where g is the image in
second figure and T = 254
17
Observations
• The isolated points in the figure are points that also had
similarly strong responses to the kernel. 1 1 1
• In the original image, these points and their immediate 1 8 1
neighbors are oriented in such a way that the kernel produced 1 1 1
a maximum response at those locations.

• These isolated points can be detected using the kernel for
point detection and then deleted, or they can be deleted using
morphological operators
18

• Three basic types of gray-level discontinuities in a digital image are:
Isolated Points
• Foreground (background) pixel surrounded by background
(Foreground) pixels
Lines
• Thin edge segment in which intensity of the background on either
side of the line is either much higher or much lower than intensity
of the line pixels
• Edges
• Edge Pixels are pixels at which intensity of image changes abruptly
19
Edge Detection
• Edge Detection
• We use edge models
• Ideally, edges should be one pixel thin
• In practice, they are blurred and noisy
• Edge models are classified according to their intensity
profiles.
A step edge is characterized by a transition

between two intensity levels occurring
ideally over the distance of one pixel
Digital step edges are used frequently as edge
Step Edge models in algorithm development
20
Edge Models
Edges that are blurred and noisy, with the degree of

blurring determined principally by limitations in the
focusing mechanism (e.g., lenses in the case of
optical images), and the noise level determined
principally by the electronic components of the
imaging system.
Ramp Edge
• The slope of the ramp is inversely proportional to the
degree to which the edge is blurred
In this model, we no longer have a single “edge

point” along the profile.
Instead, an edge point now is any point contained in
the ramp, and an edge segment would then be a set
of such points that are connected.
21
Edge Models
Roof edges are models of lines through a region, with the

base (width) of the edge being determined by the
thickness and sharpness of the line.
• In the limit, when its base is one pixel wide, a roof
edge is nothing more than a one-pixel-thick line
running through a region in an image.
Roof Edge
Roof edges arise, for example, in range imaging, when thin

objects (such as pipes) are closer to the sensor than the
background (such as walls).
The pipes appear brighter and thus create an image similar to
the model shown.
Other areas in which roof edges appear routinely are in the
digitization of line drawings and also in satellite images, where
thin features, such as roads, can be modeled by this type of
22
edge.
Edge Detection
The profiles are from dark to light, in the areas

enclosed by the small circles.
The ramp and step profiles span 9 pixels and 2
pixels, respectively. The base of the roof edge is 3
pixels.
A 1508 x 1970 image showing (zoomed) actual ramp

(bottom, left), step (top, right), and roof edge profiles.
23
Edge Detection
• Need to develop algorithms to detect these edges

• The performance of these algorithms will depend on the
differences between actual edges and the models used in
developing the algorithms
• Consider the following image and its first and second
derivatives calculated at the edge
24
Edge Detection
Detail near the edge, showing a horizontal intensity 25

profile, and its first and second derivatives.
Edge Detection
• Edge Detection
• Magnitude of the first derivative.
• Slope of first derivative is inversely proportional to degree of
blurring
• Thickness of edge is determined by length of ramp
• Sign change of the second derivative.
• Observations:
• Second derivative produces two values for an edge
(undesirable).
• Its zero crossings may be used to locate the centers of thick
edges.
26
Edge Profiles
• Profiles with embedded

noise
8-bit images with values in the range [0, 255], and

intensity profiles of a ramp edge corrupted by Gaussian
noise of zero mean and standard deviations of 0.0, 0.1,
1.0, and 10.0 intensity levels, respectively.
27
Edge Profiles
• Profiles with embedded noise

First-derivative images and intensity profiles of images
corrupted by Gaussian noise of zero mean and standard
deviations of 0.0, 0.1, 1.0, and 10.0 intensity levels,
respectively.
28
Edge Profiles
• Profiles with embedded noise

• Observations:
• Second derivative is more
sensitive to noise.
• Its zero crossings may be
used to locate the centers
of thick edges.

Second-derivative images and intensity profiles of images
corrupted by Gaussian noise of zero mean and standard
deviations of 0.0, 0.1, 1.0, and 10.0 intensity levels,
respectively. 29
Basic Edge Detection
• Detecting changes in intensity for the purpose of finding edges

can be accomplished using first- or second-order derivatives.
Image Gradient and its Properties
Gradient operation
• The tool of choice for finding edge strength and direction at an
arbitrary location (x, y) of an image, f, is the gradient, denoted
by and defined as the vector
• We can use this to find the magnitude and direction angle

• This vector has the well-known property that it points in the direction
of maximum rate of change of f at (x, y)
30
Basic Edge Detection
• Magnitude is given by:
• Direction of gradient vector is given by:
Angles are measured in the counterclockwise direction with respect to

the x-axis
This is also an image of the same size as f, created by the
elementwise division of gx and gy over all applicable values of x
and y. 31
Example
• Determine the derivatives in x and y directions using a

3x3 neighborhood centered at a point
¶f ¶f
= -2 and =2
¶x ¶y
32
Example
• Consider an image containing straight edge segment:
• Objective is to find strength and direction of edge at the

highlighted point
• Assume shaded pixel is 0 and white is 1
33
Example
• Then
• From this we get
Ñf = 2 2
34
Detection of Edge Direction
• Direction of the gradient vector at the same point is

a = tan -1 ( g y / g x ) = -450
• This is same as 1350 measured in positive

(counterclockwise) direction w.r.t x-axis
35
Detection of Edge Direction
• Direction of edge at a point is orthogonal to gradient vector at

that point
• Hence direction angle of edge is
α-900= 1350-900 =450
• All edge points in this figure have same gradient.

So entire edge segment has same direction.
• Gradient vector is sometimes called edge normal
36
Edge Linking and Boundary Detection
Local Processing
The two principal properties used for establishing similarity of
edge pixels in this kind of analysis are :
1. The strength of the response of the gradient operator used to
produce the edge pixe
2. The direction of the gradient vector
37
• An edge pixel with coordinates (x0,y0) in neighborhood of (x,y)

is similar in magnitude to the pixel at (x,y) if
where E is a positive threshold

• An edge pixel with coordinates (x0,y0) in neighborhood of (x,y)
has an angle similar to the pixel at (x,y) if
where A is a positive threshold

Direction of edge at (x,y) is perpendicular to direction of gradient
vector at that point
38
• A pixel with coordinates (x0,y0) is considered to be linked to the

pixel at (x,y) if both magnitude and angle criteria are satisfied
• This process is repeated for every edge pixel
• This is a very computationally expensive process
39
Alternate Method
• Simple procedure for real time applications is:

• Compute gradient magnitude and angle arrays of the input
image f(x,y)
• Form a binary image g(x,y) whose value at any point (x,y) is
given by
where TM is a threshold, A is a specified angle direction and ±TA

defines a band of acceptable direction about A
40
• Scan rows of g and fill (set to 1) all gaps (sets of zeros) in each
row that do not exceed a specified length, L.
• A gap is bounded at both ends by one or more 1’s
• To detect gaps on any other direction θ, rotate g by this angle
and apply the horizontal scanning procedure in Step 3. Rotate
the result by –θ
• This algorithm is more practical and gives good results
41
Edge Linking and Boundary Detection: Example
• Find rectangular shapes similar to license plate
42
Steps Involved
• Detect strong horizontal and vertical edges

• Find gradients
• Connect edge points
• Check horizontal-vertical proportion
43
44
Next Session
• Edge linking Cont..

• Thresholding
45
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
46
U3 Lec28: Image Segmentation

Dr. Shruthi M.L.J.
1
Dr. Shruthi M.L.J.

2
Last Session
•Segmentation
•Line and edge detection
•Edge Linking
3
This Session
• Segmentation
• Gradient Operators
• Thresholding
4
Gradient Operators
• Obtaining the gradient of an image requires computing the

partial derivatives ∂f /∂x and ∂f/∂y at every pixel location in
the image.
• For the gradient, we typically use a forward or centered finite
difference
• Using forward differences we obtain
1-D kernel
1-D kernel 5
Gradient Operators
• When diagonal edge direction is of interest, we need 2-D

kernels:
• Roberts operators are based on implementing the diagonal
differences
6
Gradient Operators
• Kernels of size 2 × 2 are simple conceptually, but they are not

as useful for computing edge direction as kernels that are
symmetric about their centers
• The smallest of which are of size 3 × 3.
• These kernels take into account the nature of the data on
opposite sides of the center point, and thus carry more
information regarding the direction of an edge
• The simplest digital approximations to the partial derivatives
using kernels of size 3 × 3 are given by
; approximates the derivative in the x-direction
; approximates the derivative in the y-direction

7
Gradient Operators
• Kernels used to implement these equations are
Prewitt Operators [1970]
8
Gradient Operators
• A slight variation of the preceding two equations uses a weight

of 2 in the center coefficient
Sobel Operators [1970]
9
Comparison of Operators
• The Prewitt kernels are simpler to implement than the Sobel

kernels
• Sobel kernels have better noise-suppression (smoothing)

characteristics hence more preferred
• Noise suppression is an important issue when dealing with

derivatives
10
Gradient Operators
• Any of the pairs of kernels are convolved with an image to

obtain the gradient components gx and gy at every pixel
location.
• These two partial derivative arrays are then used to estimate

edge strength and direction.
• The 3 × 3 kernels exhibit their strongest response

predominantly for vertical and horizontal edges.
11
Gradient Operators
• Obtaining the magnitude of the gradient requires the

computations of squares and square roots
• Not always desirable because of the computational burden

required by squares and square roots
• An approach used frequently is to approximate the

magnitude of the gradient by absolute values:
12
Gradient Operators
• The Kirsch compass kernels (Kirsch [1971]) are designed to detect

edge magnitude and direction (angle) in all eight compass directions
• Instead of computing the magnitude using the equations, Kirsch’s

approach was to determine the edge magnitude by convolving an
image with all eight kernels and assign the edge magnitude at a point
as the response of the kernel that gave strongest convolution value at
that point.
• The edge angle at that point is then the direction associated with
that kernel
13
Gradient Operators
If the strongest value at a point in the image

resulted from using the north (N) kernel, the edge
magnitude at that point would be assigned the
response of that kernel, and the direction would
be 0°
Kirsch compass kernels.

The edge direction of strongest response of each kernel is labeled
below it. 14
Example
Image of size 834 x 1114 pixels, the component of the gradient obtained using the Sobel operator
with in
intensity values scaled to the range the x-direction, obtained using
[0,1] the
Sobel kernel
The gradient image
15
Observations
• This example illustrates the Sobel absolute value response of the two
components of the gradient and , as well as the gradient image
formed from the sum of these two components
• The directionality of the horizontal and vertical components of the gradient
is evident in Figs. (b) and (c)
• The roof tile, horizontal brick joints and the horizontal segments of the
windows are very strong compared to other edges.
• In contrast, Fig. (c) favors features such as the vertical components of the
façade and windows.
• It is common terminology to use the term edge map when referring to an
image whose principal features are edges, such as gradient magnitude
images
16
Combining gradient with smoothing
• Edge detection can be made more selective by smoothing the

image prior to computing the gradient
Same sequence as in
earlier figure, but the
original image is smoothed
using a 5 x 5 averaging
kernel prior to edge
detection.
17
Combining gradient with thresholding
• Another approach aimed at achieving the same objective is to

threshold the gradient image.
Result of thresholding the Result of thresholding the gradient

gradient of the original of the smoothed image.
image
18
Canny Edge Detector
• Canny edge detector (Canny [1986]) is more complex but its

performance is superior in general to most edge detectors
• Canny’s approach is based on three basic objectives:
1. Low error rate:
• All edges should be found, and there should be no
spurious responses.
2. Edge points should be well localized:
• The edges located must be as close as possible to the
true edges.
• That is, the distance between a point marked as an edge by the
detector and the center of the true edge should be minimum
19
Canny Edge Detector
3. Single edge point response:

• The detector should return only one point for each true
edge point
• That is, the number of local maxima around the true edge
should be minimum.
• This means that the detector should not identify multiple edge
pixels where only a single edge point exists
• The essence of Canny’s work was in expressing the preceding

three criteria mathematically, and then attempting to find
optimal solutions to these formulations
• as achieving all three simultaneously is not possible
20
Next Session
• Edge linking using Hough Transform

• Thresholding
21
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
22
U3 Lec 29: Image Segmentation

Dr. Shruthi M.L.J.
1
Dr. Shruthi M.L.J.

2
Last Session
• Segmentation
• Gradient Operators
• Thresholding
3
This Session
• Segmentation
• Edge Linking via Hough transform
4
Canny Edge Detector
• Canny edge detector (Canny [1986]) is more complex but its

performance is superior in general to most edge detectors
• Canny’s approach is based on three basic objectives:
1. Low error rate:
• All edges should be found, and there should be no
spurious responses.
2. Edge points should be well localized:
• The edges located must be as close as possible to the
true edges.
• That is, the distance between a point marked as an edge by the
detector and the center of the true edge should be minimum
5
Canny Edge Detector
3. Single edge point response:

• The detector should return only one point for each true
edge point
• That is, the number of local maxima around the true edge
should be minimum.
• This means that the detector should not identify multiple edge
pixels where only a single edge point exists
• The essence of Canny’s work was in expressing the preceding

three criteria mathematically, and then attempting to find
optimal solutions to these formulations
• as achieving all three simultaneously is not possible
6
Global Processing
• Previous methods are applicable in situations where

knowledge about pixels belonging to individual objects is
available
• Does not work in unstructured environments in which all we
have is an edge map and no knowledge about where objects of
interest might be
• Here all pixels are candidates for linking, and thus have to
be accepted or eliminated based on predefined global
properties
• We need to develop approach based on whether sets of pixels
lie on curves of a specified shape.
• Once detected, these curves form the edges or region
boundaries of interest.
7
Global Processing
• Global Processing via the Hough Transform [1962]

• Determine if points lie on a curve of specified shape
• Unlike local analysis we now consider global relationships
between pixels
• Consider a point (xi,yi) and the general equation of a straight
b
line yi = axi +b
Slope =a
• With varying ‘a’ and ‘b’ we get infinite lines passing

through (xi,yi)
8
Global Processing using Hough Transform
• Now write the equation with respect to ab-plane

(parametric space) as b = -xi a + yi
• This gives the equation of a single line with fixed pair (xi,yi)
where the line maps to one point in xy plane
9
Global Processing using Hough Transform
• Now consider a second point (xj,yj) in the parameter space.

This intersects the line associated with (xi,yi) at (a’,b’)
• All points on a line in (x,y) plane intersect at the same point (
Ex. (aI,bI) ) in parametric plane
• In fact, all points on the line have lines in parameter space that
intersect at (a’ , b’ )
• In principle, the parameter space lines corresponding to all
points (xk , yk) in the xy-plane could be plotted, and the
principal lines in that plane could be found by identifying
points in parameter space where large numbers of parameter-
space lines intersect
10
Computational Aspects of Hough Transform
• Subdivide the parametric space into accumulator cells

• The cell at (i,j) with accumulator values A(i, j)
corresponds to (ai,bj)
• Initialize A(i, j) = 0
• For every point (xk,yk) vary a from cell to cell and solve
for b using:
• If a choice of ap results in bq , then increment the

accumulator A(p,q)=A(p,q)+1
11
Computational Aspects of Hough Transform
• At the end of the procedure, a value of Q in A(i,j)

corresponds to Q points in the xy-plane lying on the line
• K different increments of a generate K different values of b;

for n different image point, the method involves nK
computations (linear complexity)
• Number of subdivision of subdivision in ab-plane

determines the accuracy of the colinearity of these points
12
Hough Transform using Normal Representation
• Drawback of this method:

• Slope approached infinity as line approaches vertical
• Computationally very expensive
• Hence we use normal representation of line
( , ) parameterization of a line in the xy-plane

13
Hough Transform using Normal Representation
• Instead of straight lines, there are sinusoidal curves in the

parameter space
• The number of intersecting sinusoids is accumulated and then
the value Q in the accumulator A(i,j) shows the number of
collinear points lying on a line
Sinusoidal curves in the 𝜌𝜃 -plane;the point of intersection

(𝜌𝐼 , 𝜃𝐼 ) corresponds to the line passing through points (xi , Division of the 𝜌𝜃 -plane into accumulator cells.
14
yi ) and (xj , yj ) in the xy-plane.
Example
• Consider an image of size M x M (M=101) with 5 labeled

points
Image of size 101 x 101 pixels,

containing five white points
(four in the corners and one
in the center)
15
Example
• Map these points mapped onto the -plane using

subdivisions of one unit for the and axes
• The range of values is ±90°, and the range of values is

Q
100
2
R
50
Corresponding parameter space
S A 1 S
0
r
3
50 R
4
B
Q
100
80 60 40 20 0 20 40 60 80
16
u
Example
• Each curve has a different sinusoidal shape. 100

Q
• The horizontal line resulting from the mapping of point 1 is a

R
50
sinusoid of zero amplitude.

S A 1 S
0
r
3
R
• The points labeled A and B illustrate the colinearity detection

50 4
B
property of the Hough transform.

100
• For example, point B, marks the intersection of the curves

80 60 40 20 0 20 40 60 80
u
corresponding to points 2, 3, and 4 in the xy image plane.

• The location of point A indicates that these three points lie on
a straight line passing through the origin ( = 0) and oriented
at −45°
• Similarly, the curves intersecting at point B in parameter
space indicate that points 2, 3, and 4 lie on a straight line
oriented at 45°, and whose distance from the origin is = 71 17
Example
• Finally, the points labeled Q, R, and S in Figure illustrate the

fact that the Hough transform exhibits a reflective adjacency
relationship at the right and left edges of the parameter space.
• This property is the result of the manner in which and
change sign at the ±90° boundaries.
Q
100
2
R
50
S A 1 S
0
r
3
50 R
4
B
Q
100
80 60 40 20 0 20 40 60 80
18
u
Edge Linking using Hough Transform
Steps involved in Hough transform for edge linking

1. Obtain a binary edge map
2. Specify subdivisions in the -plane
3. Examine the counts of the accumulator cells for high pixel
concentrations.
4. Examine the relationship (principally for continuity)
between pixels in a chosen cell.
19
Edge Linking Using Hough Transform
• Continuity in this case usually is based on computing the

distance between disconnected pixels corresponding to a
given accumulator cell.
• A gap in a line associated with a given cell is bridged if the
length of the gap is less than a specified threshold.
• Being able to group lines based on direction is a global
concept applicable over the entire image, requiring only that
we examine pixels associated with specific accumulator cells
20
Example: Edge Linking Using Hough Transform
• Use the Hough transform to extract the two edges defining

the principal runway in the aerial image of an airport
(applications involving autonomous air navigation)
21
Step 1: Obtain the edge map
Hough parameter space (the boxes highlight the points

Edge map obtained using Canny’s algorithm associated with long vertical lines)
Obtained using 1° increments for 𝜃, and one-pixel increments
for 𝜌
22
• The runway of interest is oriented approximately 1° off the

north direction
• So we select the cells corresponding to ±90° and containing
the highest count because the runways are the longest lines
oriented in these directions.
• The small boxes on the edges of Fig below highlight these
cells.
23
• The Hough transform exhibits adjacency at the edges.

• Another way of interpreting this property is that a line
oriented at +90° and a line oriented at −90° are equivalent
(i.e., they are both vertical)
Lines in the image plane corresponding to the points highlighted by the

boxes.
24
• The lines are then superimposed on the original image.

• The lines were obtained by joining all gaps not exceeding 20%
(approximately 100 pixels) of the image height.
• These lines clearly correspond to the edges of the runway of
interest.
Lines superimposed on the original image

25
Observation
• Note that the only information needed to solve this problem was the
orientation of the runway and the observer’s position relative to it.
• In other words, a vehicle navigating autonomously would know that if the
runway of interest faces north, and the vehicle’s direction of travel also is
north, the runway should appear vertically in the image.
• Other relative orientations are handled in a similar manner.
• The orientations of runways throughout the world are available in flight
charts, and the direction of travel is easily obtainable using GPS (Global
Positioning System) information.
• This information also could be used to compute the distance between the
vehicle and the runway, thus allowing estimates of parameters such as
expected length of lines relative to image size
26
Next Session
• Segmentation by Thresholding
27
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
28

Dr. Shruthi M.L.J.
1
Dr. Shruthi M.L.J.

2
Last Session
• Segmentation
3
This Session
• Segmentation using Thresholding
4
Segmentation using Thresholding
• Thresholding enjoys a central position in applications of image

segmentation due of its intuitive properties, simplicity of
implementation, and computational speed
5
• Objective: To segment object from background

• Assume one light object and dark background
• Obtain the histogram
• Histogram dominant modes: two or more
• Threshold and thresholding operation
• Object if f(x,y)>T and background if f(x,y) <T (light object on

dark background)
6
• If there are 3 dominant modes (like 2 light objects and dark

background) : Multilevel thresholding
• A point (x,y) belongs to background if f(x,y)<T1
• To one object if T1 <f(x,y) ≤ T2
• To the other object if f(x,y) >T2
7
• The segmented image is given by:
8
Thresholding
Threshold:
• If T is a constant applicable over an entire image, it is referred
to as global thresholding
• When the value of T changes over an image, we use the term
variable thresholding.
• The terms local or regional thresholding are used
sometimes to denote variable thresholding in which the
value of T at any point (x,y) in an image depends on
properties of a neighborhood of (x,y) (for example, the
average intensity of the pixels in the neighborhood)
• If it depends on spatial coordinates x and y as well it is called
dynamic or adaptive
9
Role of illumination
• Illumination
• Image (f(x,y) is a product of reflectance and illuminance (i x r)
• Reflection depends on nature of objects and background
• Poor (nonlinear) illumination could impede the segmentation
10
• Consider a reflectance function of a noisy image and its
histogram
• Its histogram is bi-modal and can be partitioned easily:
Global T
Noisy image Corresponding histograms
11
• Now Multiplying the reflectance by illumination(non uniform)
yields the image f(x,y)=i(x,y)r(x,y)
12
• The deep valley between peaks was corrupted to the point

where separation of the modes without additional
processing is not possible
• Similar results would be obtained if the illumination was
perfectly uniform, but the reflectance of the image was
not, as a results, for example, of natural reflectivity
variations in the surface of objects and/or background.
• Illumination and reflectance play a central role in the
success of image segmentation using thresholding or other
segmentation techniques.
• Therefore, controlling these factors when possible should
be the first step considered in the solution of a
segmentation problem.
13
• There are three basic approaches to the problem when control

over these factors is not possible.
• The first is to correct the shading pattern directly.
• For example, nonuniform (but fixed) illumination can be
corrected by multiplying the image by the inverse of the
pattern, which can be obtained by imaging a flat surface of
constant intensity.
• The second is to attempt to correct the global shading pattern
via processing using, for example, the top-hat transformation
• The third approach is to “work around” nonuniformities using
variable thresholdin
14
Solution to the Illumination problem
• When the original valley is completely eliminated making

segmentation using single threshold impossible
• Solution:
f(x,y) = i(x,y)r(x,y)
Taking log of equation
Where i’ and r’ are independent random variables

• Degree of distortion would depend on broadness of
histogram of i’(x,y) which in tern depends on non
uniformity of illumination function
15
Solution to the Illumination problem
• The final histogram is a result of convolution of the histogram
of the log reflectance and log illuminance functions
• If i(x,y) was constant iI(x,y) would also be constant and its
histogram would be simple spike (like impluse).
• The convolution of this with histogram of rI(x,y) would leave its
basic shape unchanged
• But if iI(x,y) has a broader histogram (due to non uniform
illumination) the convolution would smear the histogram of
rI(x,y) yielding a histogram of z(x,y) whose shape would be
different from rI(x,y)
16
Solution
• When access to illumination source is available, a solution to
compensate for non uniformity is to project the illumination
pattern onto a constant white reflective surface
• This yields an image g(x,y)= k i(x,y), k is constant that depends
on surface
• Then for any image f(x,y) = i(x,y)r(x,y) obtained with same
illumination pattern we give normalized function
h(x,y)=f(x,y)/g(x,y)= r(x,y)/k
Thus if r(x,y) can be segmented using threshold T then h(x,y) can
be thresholded by T/k
17
Role of Noise in Image Thresholding
• Consider a synthetic image without noise and with noise
(AWGN) and their corresponding histograms
Image with AWGN of mean 0 and Image with AWGN of mean 0 and
Noiseless 8-bit image.
SD SD
of 10 intensity levels. of 50 intensity levels.
18
Corresponding Histograms
• Segmenting noiseless image into two regions is a trivial task:

we just select a threshold anywhere between the two modes.
• In the second image the modes are broader but their
separation is enough so that the depth of the valley between
them is sufficient to make the modes easy to separate.
• A threshold placed midway between the two peaks would
do the job.
• The situation in third case is much more serious, as there is no
way to differentiate between the two modes without
additional processing
19
Next Session
• Optimum Global Thresholding

• Region Based Segmentation
20
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
21
U3 Lec 31- 32: Image Segmentation

Dr. Shruthi M.L.J.
1
Dr. Shruthi M.L.J.

2
Last Session
• Segmentation
• Using Thresholding
• Role of Illumination
• Role of Noise
3
This Session

• Role of Noise Cont..
• Region based Segmentation
4
• Consider a synthetic image without noise and with noise
(AWGN) and their corresponding histograms
Image with AWGN of mean 0 and Image with AWGN of mean 0 and
Noiseless 8-bit image.
SD SD
of 10 intensity levels. of 50 intensity levels.
5
Corresponding Histograms
• Segmenting noiseless image into two regions is a trivial task:

we just select a threshold anywhere between the two modes.
• In the second image the modes are broader but their
separation is enough so that the depth of the valley between
them is sufficient to make the modes easy to separate.
• A threshold placed midway between the two peaks would
do the job.
• The situation in third case is much more serious, as there is no
way to differentiate between the two modes without
additional processing
6
Basic Global Thresholding
• Success of this method depends on how well the histogram

can be partitioned
• Consider the image shown and its histogram
7
• We can use a threshold T midway between the maximum and

minimum gray levels
• Any pixel ≤ T can be labeled black (0) and any pixel with gray
level > T can be labeled white(255)
8
• Appropriate for industrial inspection applications

with controllable illumination
• The threshold can be obtained automatically using following
algorithm
• Select an initial estimate for T
• Segment image with initial T into regions G1 and G2
• Compute the average gray level values m1 and m2 in
regions G1 and G2
• Compute new threshold T=0.5 (m1 + m2)
• Repeat until difference in T in successive iterations is
smaller than a predefined parameter T0
9
Criteria for Initial Threshold
• When the object and background occupy comparable areas in

image initial T can be average gray level of the image
• When objects are small compared to background (or vice-

versa), then one group will dominate the histogram
• Initial threshold can be midway between maximum and

minimum gray level values in such cases
10
Example
• Consider the image and its histogram
Noisy fingerprint Histogram Segmented result using a global threshold

• Iterative algorithm was used for thresholding with initial T as
average of the gray levels
• After 3 iterations threshold was 125.4 11
Example
Noisy fingerprint Segmented result using a global threshold
• Due to clear separation of modes in the histogram, the

segmentation between object and background was perfect. 12
Global Thresholding
0 63 127 191 255
Due to clear separation of

modes in the histogram,
the segmentation
between object and
background was perfect.
0 63 127 191 255
Noisy image histogram Result obtained using Otsu’s method. 13

Basic Adaptive Thresholding
• If the histogram can’t be partitioned effectively (due to some

imaging factors like uneven illumination) adaptive thresholding
can be used
• Here the image is divided into subimages and then different
thresholds(local thresholds) are used for each subimage
• Since threshold for each pixel depends on the location of the
pixel thresholding is adaptive
14
Example
• Consider the image with uneven illumination and its histogram
15
Eaxmple
• Result of global thresholding • Subdivide the image into smaller 4

taking threshold at the valley of equal parts
histogram
16
Example
• All subimages that did not contain boundary between object

and background had variance less than 75
• Can be segmented using single threshold
• Those with boundaries had variances >100
• Segmented using iterative thresholding with initial T as
midway between maximum and minimum gray levels
• Result of using adaptive threshold
17
Example
• Two subimages where boundary between object and

background was small and dark did not give good result
18
Example
• Divide the failed subimage into much smaller subimages

and use local thresholds again
• The smaller subimage has bimodal histogram and hence
easily segmentable
19
Global Thresholding
• There are various methods to improve the segmentation using

global thresholding by selecting optimum threshold:
• Otsu’s method (statistical approach)
• Image smoothing
20
Optimal Global Thresholding
• Method for estimating thresholds that produce minimum

average segmentation error
• In an image let z denote gray-level values which can be viewed
as random quantities
• Their histogram may be considered as estimate of their
probability density function(PDF),p(z)
• If image has two gray level regions then overall density
function is sum or mixture of 2 densities.
21
• If the form of densities is known or assumed, it is possible to

determine an optimal threshold (in terms of minimum error)
for segmenting the image into 2 distinct regions
• Consider PDF of 2 regions, one object(smaller) and background
(larger)
22
• Mixture probability
where P1 and P2 are probability of occurrence of 2 classes of

pixels
• Select the value of T that minimizes the average error in
making the decision that a given pixel belongs to an object or
to the background
• Probability of a random variable is integral of its PDF between
the known interval
23
• Probability of erroneously classifying a background as

object point is
• Probability of erroneously classifying an object point

as background is
24
• Then the overall probability of error is
• E1 and E2 are weighted by probability of occurrence of background

or object pixels
• To find the threshold for which error is minimum
• Differentiate the error equation, equate to zero and solve for T
• We get
P1 p1 (T ) = P2 p2 (T )
25
• This equation is solved to get optimum value of T

• If P1 = P2 then optimum threshold is where p1(z) and
p2(z) intersect
26
Thresholding based on several variables
• Thresholds based on several variables
• Color or multispectral histograms
• Thresholding is based on finding clusters in multi-
dimensional space
• Example: face detection
• Different color models
• Hue and saturation instead of RGB
27

28
Region Based Segmentation
• Segmentation: Partition an image into regions

• In all previous methods:
• Segmentation was based on finding boundaries between
regions based on discontinuities in gray levels
• Based on thresholds using distribution of pixel properties
like gray level values or color
• We can also segment image based on finding the regions
directly: Region Based Segmentation
29
• Basic formulation
• Every pixel must be in a region
• Points in a region (Ri ) must be connected
• Regions must be disjoint
• Define Logical predicate for one region and for

distinguishing between regions
30
Region Based Segmentation using Region Growing
• Region growing
• Group pixels from sub-regions to larger regions based on
predefined criteria for growth.
• Basic Approach:
• Start from a set of ’seed’ pixels and append pixels
with similar properties
• Selection of similarity criteria: color, descriptors (gray
level + moments)
• Stopping rule
31
Challenges in Region Growing
• Challenges in region based segmentation:
• The selection of similarity criteria depends not only on
the problem under consideration, but also on the type of
image data available (color/monochrome image)
• Descriptors alone can yield misleading results if
connectivity properties are not used in the region-
growing process
• Visualize a random arrangement of pixels that have three
distinct intensity values.
• Grouping pixels with the same intensity value to form a
“region,” without paying attention to connectivity, would yield a
segmentation result that is meaningless
• Formulation of a stopping rule.
• Region growth should stop when no more pixels satisfy the
criteria for inclusion in that region
32
Challenges in Region Growing
• Additional criteria that can increase the power of a region-growing

algorithm utilize the concept of size, likeness between a candidate
pixel and the pixels grown so far (such as a comparison of the intensity
of a candidate and the average intensity of the grown region)
• and the shape of the region being grown
• The use of these types of descriptors is based on the assumption that a

model of expected results is at least partially available
33
Algorithm for Region Growing
• Let f(x,y) denote an input image;
S(x,y) denote a seed array containing 1’s at the locations of seed

points and 0’s elsewhere;
and Q denote a predicate to be applied at each location (x, y).
Arrays f and S are assumed to be of the same size.
34
Algorithm for Region Growing
• A basic region-growing algorithm based on 8-connectivity may be
stated as follows:
1. Find all connected components in S(x, y) and reduce each
connected component to one pixel; label all such pixels found as 1.
All other pixels in S are labeled 0.
2. Form an image fQ such that, at each point (x, y), fQ(x, y) = 1 if the
input image satisfies a given predicate, Q, at those coordinates,
and fQ( x, y) = 0 otherwise.
3. Let g be an image formed by appending to each seed point in S all
the 1-valued points in fQ that are 8-connected to that seed point.
4. Label each connected component in g with a different region label

(e.g.,integers or letters).
This is the segmented image obtained by region growing. 35
Region Based Segmentation Example
Consider an 8-bit X-ray image of a weld (the horizontal

dark region) containing several cracks and porosities (the
bright regions running horizontally through the center of
the image).
Segment the defective weld regions using region growing
The first step is to determine the seed points.

X-Ray image of weld with
a crack that is to be
segmented We expect the regions containing these types of defects
to be significantly brighter than other parts of the X-ray
image as cracks and porosities will attenuate X-rays
considerably less than solid welds
We can extract the seed points by thresholding the

original image, using a threshold set at a high percentile
(99.9) of intensity values which is 254
36
Seed image (99.9% of max

X-Ray image of weld with value in the initial
a crack that is to be Histogram image).Crack pixels are
segmented missing.
The weld is very bright. The predicate used for region

growing is to compare the absolute difference
between a seed point and a pixel to a threshold. If the
difference is below it we accept the pixel as crack.
37
• Next, we have to specify a predicate.

• In this example, we are interested in appending to each
seed all the pixels that
(a) are 8-connected to that seed, and
(b) are “similar” to it.
Using absolute intensity differences as a measure of
similarity, the predicate applied at each location (x, y) is
38
0 63 127 191 255
Absolute value of the

Final seed image using Histogram of (e)
difference between the
morphological erosion
seed value (255) and (a)
Difference image Difference image

Segmentation result obtained
thresholded thresholded with the 39
by region growing
using dual smallest of the dual
• Region that resulted by starting with previous seeds
• Superimposing the boundaries with original image shows that

defective welds got segmented with acceptable accuracy
40
Region Splitting and Merging
• Region growing grows regions from seed points

• An alternative is to subdivide an image initially into a set of disjoint
regions and then merge and/or split the regions in an attempt to satisfy
the conditions of segmentation
Basic Approach:
• Let R represent the entire image region and select a predicate Q.
• One approach for segmenting R is to subdivide it successively into
smaller and smaller quadrant regions so that, for any region Ri ,
Q(Ri ) = TRUE
41
• We start with the entire region, R
• If Q(R) = FALSE, we divide the image into quadrants.
• If Q is FALSE for any quadrant, we subdivide that quadrant into sub-
quadrants, and so on.
• This splitting technique has a convenient representation in the form
of so-called quadtrees; that is, trees in which each node has exactly
four descendants
In this case, only R4 was subdivided further.
Partitioned Corresponding quadtree.

image. R represents the entire image region. 42
• If only splitting is used, the final partition normally contains

adjacent regions with identical properties.
• This drawback can be remedied by allowing merging as well as

splitting.
• Satisfying the constraints of segmentation requires merging only

adjacent regions whose combined pixels satisfy the predicate Q.
That is, two adjacent regions Rj and Rk are merged only if
43
Procedure for Region Splitting and Merging
1. Split into four disjoint quadrants any region Ri for which
Q(Ri ) = FALSE.
2. When no further splitting is possible, merge any adjacent

regions Rj and Rk for which
3. Stop when no further merging is possible.
44
Example: Region Splitting and Merging
• Consider a 566 × 566 X-ray image of the Cygnus Loop supernova

taken in the X-ray band by NASA’s Hubble Telescope
• Objective is to segment (extract from the image) the “ring” of

less dense matter surrounding the dense inner region
45
Observations:
• Data in this region has a random nature, indicating that its
standard deviation should be greater than the standard deviation
of the background (which is near 0) and of the large central
region, which is smooth
• The mean value (average intensity) of a region containing data
from the outer ring should be greater than the mean of the
darker background and less than the mean of the lighter central
region
• Thus, we should be able to segment the region of interest using
the following predicate:
Where and are the standard deviation and mean of
the region being processed, and a and b are nonnegative
constants.
46
Observations:
• Analysis of several regions in the outer area of interest revealed
that the mean intensity of pixels in those regions did not exceed
125, and the standard deviation was always greater than 10
• With these values for a and b vary the minimum size allowed for
the quadregions from 32 to 8.
The smallest The smallest The smallest

allowed allowed allowed
quadregion of quadregion of quadregion of 47
sizes of 32 x 32 sizes of 16 x 16 sizes of 8 x 8
• The pixels in a quadregion that satisfied the predicate were set
to white; all others in that region were set to black.
• The best result in terms of capturing the shape of the outer
region was obtained using quadregions of size 16 × 16.
• The small black squares are quadregions of size 8 × 8 whose
pixels did not satisfy the predicate.
• Using smaller quadregions would result in increasing numbers of
such black regions. Using regions larger than the one illustrated
here would result in a more “block-like” segmentation
48
• Note that in all cases the segmented region (white pixels) was a
connected region that completely separates the inner, smoother
region from the background.
• Thus, the segmentation effectively partitioned the image into
three distinct areas that correspond to the three principal
features in the image: background, a dense region, and a sparse
region.
• Using a suitable mask the white region can be separated
49
Region Segmentation Using K-Means Clustering
• This is a classical approach based on seeking clusters in

data, related to such variables as intensity and color
• Basic Idea:
• To partition a set, Q, of observations into a specified number, k, of
clusters.
• In k-means clustering, each observation is assigned to the cluster
with the nearest mean and each mean is called the prototype of its
cluster.
• A k-means algorithm is an iterative procedure that successively
refines the means until convergence is achieved
50
• Let {z1 , z2 , ... , zQ } be set of vector observations (samples).

• These vectors have the form
• In image segmentation, each component of a vector z represents a

numerical pixel attribute
• For example, if segmentation is based on just grayscale intensity, then z = z is
a scalar representing the intensity of a pixel.
• If we are segmenting RGB color images, z typically is a 3-D vector, each
component of which is the intensity of a pixel in one of the three primary
color images,
51
• The objective of k-means clustering is to partition the set Q of

observations into k (k ≤ Q) disjoint cluster sets C = {C1 , C2 , ... , Ck },
so that the following criterion of optimality is satisfied:
where mi is the mean vector (or centroid) of the samples in set Ci and
is the vector norm of the argument
• This equation says that we are interested in finding the sets
C = {C1 , C2 , ... , Ck } such that the sum of the distances from each
point in a set to the mean of that set is minimum.
• Finding this minimum is an NP-hard problem for which no practical
solution is known
• We have “Standard” k-means algorithm, based on the Euclidean
distance to find the clusters
52
Next Session
• Segmentation using watersheds
53
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
54

Dr. Shruthi M.L.J.
1
Dr. Shruthi M.L.J.

2
Last Session

• Role of Noise Cont..
• Region based Segmentation
• Region growing, splitting, merging
• Segmentation using clustering
3
This Session
• Segmentation using Morphological watershed
4

5
Segmentation Using Morphological Watersheds
• We studied segmentation using 3 principal concepts:

• Edge detection
• Thresholding and
• Region extraction
• Each of these approaches was found to have advantages and
disadvantages
• Advantages: Speed in the case of global thresholding
• Disadvantages: The need for post-processing, such as
edge linking, in edge- based segmentation
6
• Segmentation by watersheds embodies many of the concepts of the

other three approaches
• And often produces more stable segmentation results, including

connected segmentation boundaries
• This approach also provides a simple framework for incorporating

knowledge-based constraints
7
Concept: Visualize an image topographically in 3-D

• The two spatial coordinates and the intensity
• We consider three types of points
• Points belonging to a regional minimum
• Points to which a drop of water would fall certainly to
a regional minimum (catchment basin or watershed).
• Points at which the water would be equally likely to
fall to more than one regional minimum (crest lines,
divide lines or watershed lines)
Objective: The principal objective of segmentation algorithms
based on these concepts is to find the watershed lines
8
• Objective: To find the watershed lines

• Consider an example
 Topographic representation:
 The height is proportional to
the image intensity.
 Backsides of structures are
shaded for better
visualization.
Original Image Topographic View

Only the background is black.
The basin on the left is slightly lighter than black.
9
Morphological Watersheds (cont.)
• Before flooding:
• To prevent water from spilling through the image borders,
we consider that the image is surrounded by dams of
height greater than the maximum image intensity.
10
Regional minima
• Topographic representation of the image.

• A hole is punched in each regional minimum (dark areas) and
the topography is flooded by water (at equal rate) from below
through the holes.
11
• First stage of flooding.

• The water covered areas
corresponding to the dark
background.
• Next stages of flooding.
• The water has risen into the other
catchment basin.
• Further flooding.
• The water has risen into the third
catchment basin.
12
Short dam
• The water from the left basin
overflowed into the right basin.
• A short dam is constructed to
prevent water from merging.
• The effect is more pronounced.
• The first dam is now longer. New dams
• New dams are created.
Longer dam
13
• The process continues until the maximum level of flooding is

reached.
• The final dams correspond to the watershed lines which is the
result of the segmentation.
Final watershed lines

superimposed on the
image.
• Important: The watershed lines form connected paths and we

get continuous segment boundaries
14
• Summary:
• A hole is punched in each regional minimum and the topography is
flooded by water from below through the holes.
• When the rising water is about to merge in catchment basins, a dam
is built to prevent merging.
• There will be a stage where only the tops of the dams will be visible.
• These continuous and connected boundaries are the result of the
segmentation. 15
Dam Construction
• Dam construction is based on binary images, which are members of 2-D

integer space Z2
• The simplest way to construct dams separating sets of binary points is to
use morphological dilation
Flooding at stage n, showing that water has spilled

between basins
Two partially flooded catchment basins The water has spilled from one basin to the another and,
at stage n 1 of flooding therefore, a dam must be built to keep this from
happening.
16
Dam Construction Steps
Cn-1(M2)
Cn-1(M1)
Flooding step n-1. Flooding step n.
• let M1 and M2 denote the sets of coordinates of points in two regional minima.
• Then let the set of coordinates of points in the catchment basin associated with these
two minima at stage n − 1 of flooding be denoted by Cn−1(M1) and Cn−1(M2), respectively.
These are the two gray regions
• If we continue flooding, then we will have one connected component.
• This indicates that a dam must be constructed.
• Let q be the merged connected component if we perform flooding at step n.
17
• Each of the connected components is dilated by the SE shown, subject

to:
• The center of the SE has to be contained in q.
• The dilation cannot be performed on points that would cause the
sets being dilated to merge.
18
Conditions
1.Center of SE in q.
2.No dilation if merging.
• In the first dilation, condition 1 was satisfied by every point and

condition 2 did not apply to any point.
• In the second dilation, several points failed condition 1 while meeting
condition 2 (the points in the perimeter which is broken).
19
Conditions
1.Center of SE in q.
2.No dilation if merging.
• The only points in q that satisfied both conditions form the 1-pixel thick
path.
• This is the dam at step n of the flooding process.
• These points should satisfy both conditions.
20
Summary of Segmentation Techniques
• Point detection
• Line and Edge detection and linking
• Segmentation by thresholding
• Region based segmentation
• Segmentation using Morphological watersheds
21
Next Session
• Introduction to Wavelet transforms
22
THANK YOU
Dr. Shruthi M.L.J.

shruthimlj@pes.edu
+91 8147883012
23

Unit 3 (Slides) PDF

Uploaded by

Copyright:

Available Formats

Unit 3 (Slides) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Unit 3 (Slides) PDF

Uploaded by

Copyright:

Available Formats

DIGITAL IMAGE PROCESSING - 2

Lecture 24: Image Segmentation

Dr. Shruthi M.L.J.

Unit 3: Image Segmentation

Dr. Shruthi M.L.J.

• Grayscale Morphology Operations

• Image segmentation is an essential preliminary step in most

• Segmentation is to subdivide an image into constituent regions

• Image segmentation algorithms are generally based on one of

• Segmentation techniques are either region-based or edge-

• Region-based techniques rely on common patterns in

• Edge-based techniques rely on discontinuities in image

To segment the image,

Result of intensity discontinuity Result of segmentation 6

Image Segmentation Techniques

• The most common way to look for discontinuities is to run a

Point Detection Mask

Dr. Shruthi M.L.J.

Lecture 25: Image Segmentation

Dr. Shruthi M.L.J.

Unit 3: Image Segmentation

Dr. Shruthi M.L.J.

• Types of segmentation techniques

• Number of pixels available is much lower than global segmentation

• Deals with segments consisting of a relatively large number of pixels

• Segmentation techniques are either region-based or edge-

• Region-based techniques rely on common patterns in

• Edge-based techniques rely on discontinuities in image

• Derivatives of a digital function are defined in terms of finite

• Where is the separation between samples of f.

• First-order derivative at an arbitrary point x of a one-

• We compute intensity differences using just a few terms of the

• Central Difference(obtained from Taylor series by subtracting

• In general, the more terms we use from the Taylor series to

Horizontal intensity profile

• Initially, the first-order derivative is nonzero at the onset and along

• Next we encounter the isolated noise point.

• First-order derivatives generally produce thicker edges

• Second-order derivatives have a stronger response to

• Second-order derivatives produce a double-edge

• The sign of the second derivative can be used to

• The approach of choice for computing first and second

where zk is the intensity of the pixel whose spatial location

Image Segmentation Techniques

where the partial derivatives are computed using the second-

• This expression can be implemented using the Laplacian kernel

• A point has been detected at a location (x, y) on which the

where g(x, y) is the output image, T is a nonnegative threshold,

• This formulation simply measures the weighted differences

• Isolated point will be quite different from its surroundings

• Note that, as usual for a derivative kernel, the coefficients sum

• Image Segmentation Cont..

Dr. Shruthi M.L.J.

U3 Lec 26-27: Image Segmentation

Dr. Shruthi M.L.J.

Image Segmentation Techniques

Dr. Shruthi M.L.J.

Image Segmentation Techniques

where the partial derivatives are computed using the second-order