Image Processing Lecture 13

Image Segmentation
is one of image analysis methods used to subdivide an image into its
regions or objects depending on the type of shapes and objects searched
for in the image. Image segmentation is an essential first step in most
automatic pictorial pattern recognition and scene analysis tasks.

Applications of image segmentation

• Inspecting images of electronic boards for missing components or
broken connections.
• Detecting faces, facial features and other objects for surveillance.
• Detecting certain cellular objects in biomedical images.

Segmentation Approaches
Image segmentation algorithms are based on one of two basic properties
of gray-level values: discontinuity and similarity.
• In the first category, the approach is to partition an image based on
abrupt discontinuity (i.e. change) in gray level, such as edges in an
image.
• In the second category, the approaches are based on partitioning an
image into regions that are similar according to a set of predefined
criteria.

We shall focus on segmentation algorithms to detect discontinuities such

as points, lines and edges. Segmentation methods, studied here, rely on
two steps:

©Asst. Lec. Wasseem Nahy Ibrahem Page 1

Image Processing Lecture 13

1. Choosing appropriate filters that help highlight the required

feature(s).
2. Thresholding.

Point Detection
This is concerned with detecting isolated image points in relation to its
neighborhood which is an area of nearly constant gray level.

1. Simple method
The simplest point detection method works in two steps:
1. Filter the image with the mask:
-1 -1 -1
-1 8 -1
-1 -1 -1

Then, we take the absolute values of the filtered image.

2. On the filtered image apply an appropriate threshold (e.g. the
maximum pixel value).

The next figure shows an example of point detection in a face image

using the simple method.

©Asst. Lec. Wasseem Nahy Ibrahem Page 2

Image Processing Lecture 13

(a)

(b) Result with (c) Result with (d) Result with

threshold=max threshold=220 threshold=168

(e) Result with (f) Result with (g) Result with

threshold=118 threshold=68 threshold=55

Figure 13.1 Example of point detection using simple method. (a) Original face image.
(b)-(g) Results with different Thresholds

©Asst. Lec. Wasseem Nahy Ibrahem Page 3

Image Processing Lecture 13

2. Alternative method
An alternative approach to the simple method is to locate the points in a
window of a given size where the difference between the max and the
min value in the window exceeds a given threshold. This can be done
again in two steps:
1. Obtain the difference between the max value (obtained with the
order statistics max filter) and the min value (obtained with the
order statistics min filter) in the given size mask.
2. On the output image apply an appropriate threshold (e.g. the
maximum pixel value).
The figure below shows an example of point detection in a face image
using the alternative method.

(b) Threshold=max (c) Threshold=90

(a)

(d) Threshold=40 (e) Threshold=30

Figure 13.2 Example of point detection using alternative method. (a) Original face image.
(b)-(e) Results with different Thresholds

©Asst. Lec. Wasseem Nahy Ibrahem Page 4

Image Processing Lecture 13

Line Detection
Detecting a line in a certain direction require detecting adjacent points in
the image in the given direction. This can be done using filters that yields
significant response at points aligned in the given direction.
For example, the following filters
-1 2 -1 -1 -1 -1
-1 2 -1 2 2 2
-1 2 -1 -1 -1 -1

-1 -1 2 2 -1 -1
-1 2 -1 -1 2 -1
2 -1 -1 -1 -1 2

highlight lines in the vertical, horizontal, +45° direction , and – 45°

direction, respectively.
This can be done again in two steps:
1. Filter the image using an appropriate filter.
2. Apply an appropriate threshold (e.g. max value).

The next figure illustrates an example of line detection using the filters
above.

©Asst. Lec. Wasseem Nahy Ibrahem Page 5

Image Processing Lecture 13

(a)

(b) (c)

(d) (e)
Figure 13.3 Example of line detection. (a) Original image. (b)-(e) Detected lines in the
vertical, horizontal, +45° direction , and – 45° direction, respectively.

©Asst. Lec. Wasseem Nahy Ibrahem Page 6

Image Processing Lecture 13

Edge detection
Edge detection in images aims to extract meaningful discontinuity in
pixel gray level values. Such discontinuities can be deduced from first
and second derivatives as defined in Laplacian filter.
The 1st-order derivative of an image f(x,y) is defined as:

⎡ ⎤
⎢ ⎥
∇ = =⎢ ⎥
⎢ ⎥
⎣ ⎦
Its magnitude is defined as:

∇ = +

Or by using the absolute values

∇ ≈| |+

The 2nd-order derivative is computed using the Laplacian as follows:

( , ) ( , )
∇ = +

However, Laplacian filter is not used for edge detection because, as a

second order derivative:
• it is sensitive to noise.
• its magnitude produces double edges.
• it is unable to detect edge direction.

There are 1st-order derivative estimators in which we can specify whether

the edge detector is sensitive to horizontal or vertical edges or both. We
study only two edge detectors namely Sobel and Prewitt edge detectors.

©Asst. Lec. Wasseem Nahy Ibrahem Page 7

Image Processing Lecture 13

Sobel edge detector

This detector uses the following masks to approximate the digitally the
1st-order derivatives Gx and Gy:
-1 -2 -1 -1 0 1
0 0 0 -2 0 2
1 2 1 -1 0 1

To detect:
• Horizontal edges, we filter the image f using the left mask above.
• Vertical edges, we filter the image f using the right mask above.
• Edges in both directions, we do the following:
1. Filter the image f with the left mask to obtain Gx
2. Filter the image f again with the right mask to obtain Gy
3. Compute = + or ≈| |+
In all cases, we then take the absolute values of the filtered image, then
apply an appropriate threshold.

The next figure shows an example of edge detection using the Sobel
detector.

©Asst. Lec. Wasseem Nahy Ibrahem Page 8

Image Processing Lecture 13

(a) (b)

(c) (d)
Figure 13.4 Example of Sobel edge detection. (a) Original image.
(b)-(d) Edges detected in vertical, horizontal, and both directions, respectively.

Prewitt edge detector.

This detector uses the following masks to approximate Gx and Gy:
-1 -1 -1 -1 0 1
0 0 0 -1 0 1
1 1 1 -1 0 1
The steps of applying this detector are the same as that of the Sobel
detector.
The next figure shows an example of edge detection using the Prewitt
detector.

©Asst. Lec. Wasseem Nahy Ibrahem Page 9

Image Processing Lecture 13

(a) (b)

(c) (d)
Figure 13.5 Example of Prewitt edge detection. (a) Original image.
(b)-(d) Edges detected in vertical, horizontal, and both directions, respectively.

We can see that the Prewitt detector produces noisier results than the
Sobel detector. This is because the coefficient with value 2 in the Sobel
detector provides smoothing.

©Asst. Lec. Wasseem Nahy Ibrahem Page 10