International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

Image Segmentation using Discontinuity-Based Approach

Rajiv Kumar M.Arthanari M.Sivakumar

Bharathiar University, Bharathidasan School of Anna University, Coimbatore
Coimbatore (Tamil Nadu) Computer Applications, (Tamil Nadu)
Erode (Tamil Nadu)

Abstract techniques for monochrome image segmentation. In

Image segmentation is the process by which we discontinuity-based approach, the partitions or sub-
segment a given image into several parts so that we division of an image is based on some abrupt changes
can further analyzed each of these parts present in in the intensity level of images. Here, we mainly
the image. We can extract some information by interest in identification of isolated points, lined and
analyzing them and this information is useful for edges in an image.
high-level machine vision application. There are Section II describes the related work in the area of
numerous techniques of image segmentation edge detection algorithms. Section III discusses the
available in literature. In this paper, we will analyze image segmentation based on discontinuity-based
the discontinuity-based approach for image approach. Under this approach, we analyses the point
segmentation. The discontinuity-based segmentation detection, line detection and edge detection
can be classified into three approaches: point techniques. A number of operator which are based on
detection, line detection, and edge detection. In this first-order derivatives and second-order derivatives
paper, we will implement and analyses the result of such as Prewitt, Sobel, Roberts etc. are discussed in
these various approaches in MATLAB using Image detail. Section IV performs the experimental results
Processing Toolbox (IPT). We will also implement and discussion based on these operators. Finally,
the different edge operators such as Prewitt, Roberts, section V concludes the papers by analyzing and
LoG, Canny and the results of these operators will be comparing the results of these operators.
shown on various images.
2. Related work
1. Introduction
A. Rosenfeld [1] develop a method for detecting
Image segmentation is the process of dividing an edges between regions of different average gray level
image into different regions such that each region is can be applied to detect a wide variety of “texture
homogeneous. Image segmentation is the key behind edges” in which two regions differ with respect to the
image understanding. Image segmentation is average value of some local property. E. Adward [2]
considered as an important basic operation for used a convolution method to find the efficient edge
meaningful analysis and interpretation of image detector with an adjustable parameter k that may be
acquired. It is a critical and essential component of an selected to match the average interval between edges
image analysis and/or pattern recognition system, and expected in the data. The Gaussian function (G) has
is one of the most difficult tasks in image processing, only one discontinuity and therefore generates less
which determines the quality of the final noise than the other convolving function. R. M.
segmentation. It is the prime area of research in Haralick [3] uses the facet model to accomplish step
computer vision. A number of image segmentation edge detection. The essence of the facet model is that
techniques are available, but there is no one single any analysis made on the basis of the pixel values in
technique that is suitable to all the application. some neighborhood has its final authoritative
Researchers have extensively worked over this interpretation relative to the underlying gray tone
fundamental problem and proposed various methods intensity surface of which the neighborhood pixel
for image segmentation. These methods can be values are observed noise samples. He define an edge
broadly classified into seven groups: (1) Histogram to occur in a pixel if and only if there is some point in
thresholding, (2) Clustering (Fuzzy and Hard), (3) the pixel’s area having a negatively sloped zero
Region growing, region splitting and merging, (4) crossing of the second directional derivative taken in
Discontinuity-based, (5) Physical model- based, (6) the direction of a nonzero gradient at the pixel’s
Fuzzy approaches, and (7) Neural network and GA center. J. Canny [4] describes a computational
(Genetic algorithm) based approaches. Discontinuity- approach to edge detection. He defines detection and
based segmentation is one of the widely used localization criteria for a class of edges, and present

Copyright © 2011, Infonomics Society 72

International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

mathematical forms for these criteria as functional on evaluation of image segmentation systematically;
the operator impulse response. A third criterion is evaluate and compare basic, practical segmentation
then added to ensure that the detector has only one algorithms after a large number of comparative
response to a single edge. N. R. Pal and S. K. Pal [5] experiments. R. Maini et al. [14] analyzed and did
reviews and summarizes some image segmentation the visual comparison of the most commonly used
algorithms which are based on fuzzy set, neural Gradient and Laplacian based edge detection
networks, thresholding, edge detection, clustering techniques for problems of inaccurate edge detection,
relaxation and Markov Random Field (MRF). They missing true edges, producing thin or thick lines and
also address the issue of quantitative evaluation of problems due to noise etc. P. Zhou et al. [15] discuss
segmentation results. T. Kanungo et al. [6] outline a the basic theory of edge detection; its method based
methodology for summarizing many operating curves on the traditional canny operator, and proposes an
into a few performance curves. Their methodology is improved algorithm based on the eight neighborhood
adapted from the human psychophysics literature and gradient magnitude to overcome the disadvantages of
is general to any detection algorithm. The central being sensitive to noise in the calculation of the
concept is to measure the effect of variables in terms traditional canny operator gradient. The experimental
of the equivalent effect of a critical signal variable, results prove that this improved method can
which in turn facilitates the determination of the effectively detect the edge of the image, and the
breakdown point of the algorithm. They demonstrate continuity of the edge is strong, and positioning
the methodology by comparing the performance of accuracy is high.
two-line detection algorithms. W. Y. Ma et al. [7]
develop a framework for detecting and integrating 3. Discontinuity-based segmentation
intensity/color and texture discontinuities as well as
illusory boundaries in images. They propose an edge In discontinuity-based approach, the partitions or
flow model for boundary detection and image sub-division of an image is based on some abrupt
segmentation. Salient features of this approach changes in the intensity level of images. Here, we
include (1) use of a predictive coding model for mainly interest in identification of isolated points,
identifying and integrating the different types of lined and edges in an image. To identify these, we
image boundaries, (2) boundary detection based on use 3 3 Mask operation.
flow field propagation, and (3) very few “free”
parameters that control the segmentation. C.
Hampton et al. [8] summarized the various image
segmentation techniques based on discontinuity-
based, region-based, thresholding and histogram. K.
S. Fu et al. [9] summarized some image segmentation
techniques. These techniques can be categorized into
three classes, (1) characteristic features thresholding
or clustering, (2) edge detection, and (3) region
extraction. N. Senthikumaran et al. [10] survey the
theory of edge detection for image segmentation (a)
using soft computing approach based on the Fuzzy
logic, Genetic Algorithm and Neural Network. S. S.
Al-amri et al. [11] presented methods for edge
segmentation of satellite image: they used seven
techniques for this category; Sobel operator
technique, Prewitt technique, Kiresh technique,
Roberts technique, Laplacian technique, Canny
technique and Edge Maximization Technique (EMT) (b)
and they are compared with one another so as to Figure 1. (a) An image (b) 3 3 Mask
choose the best technique for edge detection segment
image. P. Thakare [12] discussed about some image
segmentation techniques like edge based, region
based and integrated techniques and explains in brief
the edge based techniques and their evaluation. They  
also focuses on edge based techniques and their
evaluation. W. Kang et al. [13] classify and discuss
main image segmentation algorithms; introduce the

Copyright © 2011, Infonomics Society 73

International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

The discontinuity-based segmentation can be

classified into three approaches: (1) Point detection, -1 -1 -1 -1 -1 2
(2) Line detection, and (3) Edge detection.

3.1. Point Detection [8] [16] 2 2 2 -1 2 -1

A point is the most basic type of discontinuity in a

digital image. The most common approach to finding -1 -1 -1 2 -1 -1
discontinuities is to run an (n n) mask over each
point in the image. The mask is as shown in figure 2.
(a) (b) 
-1 -1 -1

-1 2 -1 2 -1 -1
-1 8 -1

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

Figure 2. A mask for point detection

-1 2 -1 -1 -1 2
The point is detected at a location (x, y) in an image
where the mask is centered. If the corresponding
value of R such that (c) (d)

Where R is the response of the mask at any point in Figure 4. Line Detector masks in (a) Horizontal
the image and T is non-negative threshold value. It direction (b) 45° direction (c) Vertical direction
means that isola ted point is detected at the (d) - 45° direction
corresponding value (x, y). This formulation serves to
measures the weighted differences between the center The greatest response calculation from these matrices
point and its neighbors since the gray level of an will yield the direction of the given pixel []. The
isolated point will be very different from that of its result of line detection mask is as shown in figure 5.
neighbors [ ]. The result of point detection mask is as
shown in figure 3.

Figure 3. (a) Gray-scale image with a nearly

invisible isolated black point (b) Image showing
the detected point

3.2. Line Detection [8] [16]

Figure 5. (a) Original Image (b) result showing
Line detection is the next level of complexity in with horizontal detector (c) with 45° detector (d)
the direction of image discontinuity. For any point in with vertical detector (e) with -45° detector
the image, a response can be calculated that will
show which direction the point of a line is most With the help of lines detector masks, we can detect
associated with. For line detection, we use two the lines in a specified direction. For example, we are
masks, and, mask. Then, we have interesting in finding all the lines that are one pixel
thick, oriented at -45°. For that, we take a digitized
(binary portion of a wire-bond mask for an
It means that the corresponding points is more likely
electronics circuit. The results are shown as in figure
to be associated with a line in the direction of the
6.
mask i.

Copyright © 2011, Infonomics Society 74

International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

3.3.1. First-order derivatives. First-order

derivatives responds whenever there is discontinuity
in intensity level. It is positive at the leading edge and
negative at the trailing edge. Suppose we have an
image f(x, y) and gradient operator f.

Strength of is given by

Figure 6. (a) Image of a wire-bond mask (b)

Result of processing with the -45° detector (c)
Zoomed view of the top, left region of -45°
detector (d) Zoomed view of the bottom, right It gives the strength of edge at location (x, y)
region of -45° detector (e) Absolute value of -45° Direction of is given by
detector (f) All points whose values satisfied the
condition g >= T, where g is the image in (e)

3.3. Edge detection [8] [16] Where gives the direction of .

Since isolated points and lines of unitary pixel

thickness are infrequent in most practical application,
edge detection is the most common approach in gray
level discontinuity segmentation. An edge is a
boundary between two regions having distinct
intensity level. It is very useful in detecting of
discontinuity in an image. When the image changes
from dark to white or vice-versa. The changes of (a) (b) (c) (d)
intensity, first-order derivative and second-order Figure 8. (a) Original Image (b) component
derivative are shown in figure 7. of the gradient along x-direction (c)
Dark White Dark White Dark White component of the gradient along y-direction (d)
Gradient Image
There is several ways to calculate the image gradient:

3.3.1.1. Prewitt Edge operator

(a)
-1 -1 -1 -1 0 -1

(b) -1 0 -1
0 0 0

-1 -1 -1 -1 0 -1
(c)

Figure 9. Masks used for Prewitt Edge operator

The mask finds the horizontal edges is equivalent to

gradient in the vertical direction and the mask
compute the vertical edges is equivalent to gradient
in the horizontal direction. Using these two masks
Figure 7. (a) Intensity profile (b) First-order passing to the intensity image, we can find out and
derivatives (c) Second-order derivatives component at different location in an image. So,

Copyright © 2011, Infonomics Society 75

International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

we can find out the strength and direction of edge at

that particular location (x, y).
-1 -1 -1 1 1 1

-1 8 -1 1 8 1
3.3.1.2. Sobel Edge operator

-1 0 1 -1 -1 -1 1 1 1
-1 -2 -1
Figure 12. Masks used for Laplacian operator
0 0 0 -2 0 2 using 8-connectivity

It is not used for edge detection because it is very

1 2 1 -1 0 1 sensitive to noise and also leads to double edge. But,
it is very useful for extracting secondary information.
Figure 10. Masks used for Sobel Edge operator To reduce the effect of noise, first image will be
smooth using the Gaussian operator and then it is
It gives the averaging effect over an image. It operated by Laplacian operator. These two operations
considers the effect due to the spurious noise in the together is called LoG (Laplacian of Gaussian)
image. It is preferable over prewitt edge operator operator.
because it gives the smoothing effect and by which
we can reduce spurious edge which are generated 3.3.2.2. LoG operator
because of noise present in the image.
The LoG mask is given by
3.3.2. Second-order derivatives
0 0 -1 0 0
It is positive at the darker side and negative at the
white side. It is very sensitive to noise present in an
image. That’s why it is not used for edge detection. 0 -1 2 -1 0
But, it is very useful for extracting some secondary
information i.e. we can find out whether the point lies
on the darker side or the white side. -1 -2 1 -2 -1
Zero-crossing: It is useful to identify the exact
location of the edge where there is gradual transition 0 -1 -2 -1 0
of intensity from dark to bright region and vice-versa.
There are several second-order derivative operators:
0 0 -1 0 0
3.3.2.1. Laplacian operator. The Laplacian mask is
given by:
  Figure 13. Masks used for LoG operator
 
0 -1 0 The Gaussian operator is given by:
 
 
-1 4 -1
 
 
0 -1 0

Figure 11. Masks used for Laplacian operator

3.3.2.3. Canny operator [4]

It is very important method to find edges by isolating

noise from the image before find edges of images,
If we consider the diagonal elements: without affecting the features of the edges in the
image and then applying the tendency to find the
edges in the image and the critical value for
threshold.

Copyright © 2011, Infonomics Society 76

International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

i. Convolve image f(r, c) with a Gaussian edge is weak, while for Sobel’s operator it is not that
function to get smooth image g(r, c). weak as it gives greater weights to points lying close
f(r, c) = f(r, c) * G(r, c) to the point (x, y) under consideration. However, both
ii. Apply first difference gradient operator to Prewitt’s and Sobel’s operators posses greater noise
compute edge strength. immunity. The preceding operators are called the first
iii. Apply non-maximal or critical suppression difference operator. Laplacian, on the other hand, is a
to the gradient magnitude. second difference operator. A good edge detector
iv. Apply threshold to the non-maximal should be a filter with the following two features.
suppression image. First, it should be a differential operator, taking either
a first or second spatial derivative of the image.
4. Experimental results and discussion Second, it should be capable of being tuned to act at
any desired scale, so that large filters can be used to
The various edge detectors are implemented in detect sharply focused fine details. The second
MATLAB using Image Processing Toolbox (IPT) requirement is very useful as intensity changes occur
and the result are shown as in figure 15. at different scales in an image. According to N. R.
Pal [5], the most satisfactory operator fulfilling these
conditions is the Laplacian of Gaussian (LoG)
operator. The Gaussian part of the LoG operator
blurs the image, wiping out all structures at scales
much smaller than the of the Gaussian. The
Gaussian blurring function is preferred over others
because it has the desirable property of being smooth
and localized in both spatial and frequency domains.
According to canny [4], a good edge detector should
have the following three properties: (1) low
(a) (b)
probability of wrongly marking non-edge points and
low probability of failing to mark real edge points
(i.e. good detection); (2) points marked as edges
should be as close as possible to the centre of true
edges (i.e. good localization); and (3) one and only
one response to a single edge points (single
response). Good detection can be achieved by
maximizing signal to noise ratio (SNR), while for
good localization; Canny used the reciprocal of an
estimate of the r.m.s. distance of the marked edge
(c) (d) from the center of the true edge. To maximize
simultaneously both good detection and localization
criteria, he maximized the product of SNR and the
reciprocal of standard deviation (approximate) of the
displacement of edge points. The maximization of the
product is done subject to a constraint which
eliminates multiple responses to a single edge points.

5. Conclusion
(e) (f)
Figure 14. (a) Original image (b) Result using with In this paper, we analysis various image
Prewitt operator (c) Result using with Roberts segmentation techniques based on the discontinuity-
operator (d) Result using with Sobel operator (e) based approach. In point-detection method, the point
Result using with LoG operator (f) Result using is detected at a location (x, y) in an image where the
with Canny operator mask is centered. In line detection method, we have
two masks so that the corresponding points are more
These different operators respond to changes in gray likely to be associated with a line in the direction of
level or average gray level. The gradient operators, the one mask as compare to the one. In edge
not only respond to edges but also to isolates points. detection method, we analyses various operators such
For Prewitt’s operator the response to the diagonal as Prewitt, Roberts, Sobel, LoG and canny. These

Copyright © 2011, Infonomics Society 77

International Journal Multimedia and Image Processing (IJMIP), Volume 1, Issues 3/4, September/December 2011

different edge operators are implemented in matlab [12] Puman Thakare, “A study of image segmentation and
using image processing toolbox (IPT). The Prewitt’s edge detection techniques”, International Journal of
operator gives the weak response to the diagonal Computer Science and Engineering (IJCSE), vol. 3, no. 2,
edge while Sobel operator gives the strong response. Feb 2011.
However, both these operators posses greater noise [13] W. Kang, Q. Yang and R. Liang, “The Comparative
immunity. Laplacian of Gaussian (LoG) operator Research on Image Segmentation Algorithms”.
gives much better result as compare to Laplacian and
Gaussian operator. In LoG, intensity changes occur at [14] R. Maini and H. Aggarwal, “study and Comparsion of
different scales in an image. The canny edge detector various Image Edge Detection Techniques”, International
gives the better results as compare to other edge Journal of Image Processing (IJIP), vol. 3, issue 1.
detectors because it provide good detection, good
localization and single response. [15] P. Zhou, W. Ye, Y. Xia, Q. Wang, “An Improved
Canny Algorithm for Edge Detection”, Journal of
Computational Information System 7:5 (2011) 1516-1523.
6. References
[16] R. Gonzalez and R. Woods. Digital Image Processing.
[1] A. Rosenfeld, “A Nonlinear Edge Detection Addision-Wesley Publishing Company, Reading, MA,
Techniques”, Proceedings of the IEEE, May 1970 1993.

[2] E. Argyle, “Techniques for edge detection”, Proc.

IEEE, vol. 59, pp. 285-286, 1971.

[3] R. M. Haralick, “Digital Step Edges from Zero Crossing

of Second Directional Derivatives”, IEEE Transactions on  
Pattern Analysis and Machine Intelligence, vol. Pami-6,
no. 1, Jan. 1986.

[4] J. F. Canny, “A computational approach to edge

detection”, IEEE Trans. Pattern Anal. Machine Intell., vol.
PAMI-8, no. 6, pp. 679-697, 1986.
 
[5] N. R. Pal and S. K. Pal, “A Review on Image
Segmentation Techniques”, Pattern Recognition, vol. 26,
no. 9, pp. 1277-1294, 1993.

[6] T. Kanungo, “A Methodology for Quantitative

Performance Evaluation of Detection Algorithms”, IEEE
Transactions on Image Processing, vol. 4, no. 12, Dec.
1995.

[7] W. Y. Ma and B. S. Manjunath, “Edge Flow: A

Framework of Boundary Detection and Image
Segmentation”, IEEE, 1997.
http://www.vivaldi.ece.ussb.edu. 

[8] C. Hampton, T. Persons, C. Wyatt and Y. Zhang,

“Survey of Image Segmentation”, April 23, 1998.

[9] K. S. Fu, “A survey on image segmentation”, Pattern

Recognition, vol. 13, pp. 3-16, 1981.

[10] N. Senthikumaran and R. Rajesh, “Edge detection

techniques for image segmentation – A survey”,
Proceedings for the International Conference on Managing
Next Generation Software Applications (MNGSA-08),
2008, pp. 749-760.

[11] S. S. Al-amri, N. V. Kalyankar and S. D. Khamitkar,

“Image Segmentation by using Edge Detection”,
International Journal on Computer Science and
Engineering (IJCSE), vol. 02, no. 03, pp. 804-807, 2010.

Copyright © 2011, Infonomics Society 78