International Journal of Engineering Innovation & Research

Volume 1, Issue 1, ISSN 2277 – 5668

 

Analysis of Image Segmentation Algorithms

 
Using MATLAB
Gajendra Singh Chandel Ravindra Kumar Deepika Khare Sumita Verma
[email protected] [email protected] [email protected] [email protected]

Abstract— Image segmentation has played an important Fig.1 indicates the classification of image segmentation
role in computer vision especially for human tracking. The techniques we have considered in this paper. The methods
result of image segmentation is a set of segments that explained and used to segment the image in fig.3, fig.4,
collectively cover the entire image or a set of contours fig.5, fig.6 and fig.7 were used only to clarify the
extracted from the image. Its accuracy but very elusive is segmentation methods.
very crucial in areas as medical, remote sensing and image
retrieval where it may contribute to save, sustain and protect
human life. This paper presents the analysis and
implementation using MATLAB features and one best result
can be selected for any algorithm using the subjective
evaluation. We considered the techniques under the following
five groups: Edge-based, Clustering-based, Region-based,
Threshold-based and Graph-based.

Key Words— Image segmentation, N-cut, Mean-shift,

Fuzzy-C mean, Image analysis.

I. INTRODUCTION
The main goal of image segmentation is domain
independent partitioning of an image into a set of disjoint
regions that are visually different, homogeneous and
meaningful with respect to some characteristics or
computed property(ies), such as grey level, texture or
color to enable easy image analysis (object identification,
classification and processing).
Discontinuity and similarity/homogeneity are two basic
properties of the pixels in relation to their local
neighborhood used in many segmentation methods. The Fig.1 classification of image segmentation techniques
segmentation methods that are based on discontinuity
property of pixels are considered as boundary or edges II. MATLAB
based techniques and that are based on similarity or
homogeneity are region based techniques. Unfortunately, Matlab (MATrix LABoratory) is a tool to do numerical
both techniques often fail to produce accurate computations, display information graphically in 2D and
segmentation results [2]. Image segmentation is used in 3D, and solve many other problems in engineering and
various applications. For all the applications, a single science. Developed by MathWorks, MATLAB allows
method cannot produce the desired result. It is all due to matrix manipulations, plotting of functions and data,
that the images have different property and some other implementation of algorithms, creation of user interfaces,
factors also like noise, brightness etc. put impact on the and interfacing with programs written in other languages,
images, and it is not possible to apply a single including C, C++, Java, and Fortran[3].
segmentation method and also a single evaluation Matlab is an interpreted language for numerical
technique for all types of imagery. computation. It allows one to perform numerical
This paper analyzes the results of various segmentation calculations, and visualize the results without the need for
algorithms, using the subjective evaluation, on the complicated and time consuming programming. Matlab
different types of images and particularly on gray level allows its users to accurately solve problems, produce
images. graphics easily and produce code efficiently [4].
In this paper, a program has been developed using
This paper will be organized as follows: MA TLAB to load the images, which contained link for all
• MATLAB algorithms using pushbutton, pop-up menus and sliders to
• Segmentation Results change the values of the parameters related to the
• Implementation of the proposed system concerned method such as number of regions in region-
• Performance Evaluation
• Conclusion.
Copyright © 2012 IJEIR, All right reserved 51
 International Journal of Engineering Innovation & Research
Volume 1, Issue 1, ISSN 2277 – 5668
 

based methods value of threshold in threshold-based which maximizes this function is chosen as the threshold.
methods. For Example:

III. SEGMENTATION RESULT

A. Edge-based Segmentation Algorithms

Edge-base segmentation generally indicates the
segmentation method based on the edge in an image. The
simple methods apply some edge detection methods before 2. Otsu Thresholding Method
segmentation. Otsu's thresholding method involves iterating through
For Example: all the possible threshold values and calculating a
measure of spread for the pixel levels each side of the
threshold, i.e. the pixels that either fall in foreground
or background. The aim is to find the threshold value
ORIGINAL IMAGE where the sum of foreground and background spreads is at
its minimum.
For Example:

PREWITT
C. Region-Based Segmentation Algorithms
Region-based methods mainly rely on the assumption
that the neighboring pixels within one region have similar
value. It compare one pixel with its neighbors. If a
similarity criterion is satisfied, the pixel can be set belong
to the cluster as one or more of its neighbors. The
ROBERTS
selection of the similarity criterion is significant and the
results are influenced by noise in all instances. Two
algorithms are used:
1. Seeded Region Growing Method
Step1: We start with a number of seed points which have
been clustered into n clusters and the positions of initial
LAPLACIAN of seed points.
GAUSSIAN (LoG) Step2: Recompute the boundary of and set those boundary
points as new seed points.
Step3: Repeat Step2 until all pixels in image have been
allocated to a suitable cluster.
For Example:

CANNY

B. Thresholding Based Segmentation Algorithm 2. Region Splitting and Merging Method

Thresholding becomes then a simple but effective tool to Step1: Splitting steps: For any region Ri, split it into four
separate objects from the background. The output of the disjoint quadrants.
thresholding operation is a binary image whose gray level Step2: Merging steps: When no further splitting is
of 0 (black) will indicate a pixel belonging to a print, possible, merge any adjacent regions Rj and Rk.
legend, drawing, or target and a gray level of 1 (white) Step3: Stop only if no further merging is possible.
will indicate the background. Two algorithms are used: For Example:
1. Adaptive Thresholding Method
In adaptive thresholding, a criterion function is devised
that yields some measure of separation between regions. A
criterion function is calculated for each intensity and that
Copyright © 2012 IJEIR, All right reserved 52
 International Journal of Engineering Innovation & Research
Volume 1, Issue 1, ISSN 2277 – 5668
 

the cluster centers in the feature domain.

For example:

D. Clustering Based Segmentation Methods

Clustering is one of methods widely applied in image
segmentation and statistic. The main concept of clustering
is to use the centroid to represent each cluster and base on E. Graph Based Segmentation Algorithm
the similarity with the centroid of cluster to classify. The graph based image segmentation is based on
According to the characteristics of clustering algorithm, selecting edges from a graph, where each pixel
we can roughly divide into “hierarchical (i.e, nested corresponds to a node in the graph. Weights on each edge
grouping of patterns )” and “partitioned (i.e, select the measure the dissimilarity between pixels. The
number of desired output clusters )” clustering. Three segmentation algorithm defines the boundaries between
algorithms are used: regions by comparing two quantities – Intensity
1. Mean Shift Method differences across the boundary and Intensity difference
Step 1: Determine the number of clusters we want in the between neighboring pixels within each region.
final classified result and set the number as N.
Step2: Classify each pattern to the closest cluster centroid. 1. Normalized-Cut Method
The closest usually represent the pixel value is similarity, Step1:The input is a graph
but it still can consider other features. G = (V,E),
Step3: Recompute the cluster centroids and then there where V are the n vertices and E are m edges. Each
have N centroids of N clusters as we do after Step1. edge has a corresponding weight.
Step4: Repeat the iteration of Step 2 to 3 until a Step 2: Perform the segmentation.
convergence criterion(no reassignment of any pattern from Step 3: If the weight of the edge connecting two
one cluster to another, or the minimal decrease in squared vertices in adjacent components is small compared to
error) is met. the internal difference of both the components, then
For example: merge the two components, otherwise do nothing.
Step 4: Repeat Step3.
For Example:

2. K-Means Clustering Method

Choose k-data points to act as cluster centers Until the
cluster centers are unchanged. Allocate each data point to
cluster whose center is nearest. Now ensure that every
cluster has at least one data point; possible techniques for
doing this include supplying empty clusters with a point IV. IMPLEMENTATION OF THE PROPOSED
chosen at random from points far from their cluster center. SYSTEM
Replace the cluster centers with the mean of the  
elements in their clusters. • Initially an image is selected and then converted into
For Example:
Binary and Gray level image of size 256x256.
• The original and resized image is displayed.
• The value of the concerned parameters, if any, is
given.
• The result of each algorithm is displayed in the
separate figure window.
3. Fuzzy C-Means Clustering Method • The segmented image is displayed at a particular
An image can be represented in various feature spaces, position according to the range of the selected value
and the FCM algorithm classifies the image by grouping of the parameter.
similar data points in the feature space into clusters. This Above step is repeated with various values of
clustering is achieved by iteratively minimizing a cost parameters for the same algorithm and the results are
function that is dependent on the distance of the pixels to obtained in the same figure window.

Copyright © 2012 IJEIR, All right reserved 53

 International Journal of Engineering Innovation & Research
Volume 1, Issue 1, ISSN 2277 – 5668
 

With the help of those results in a figure window, a best It is very sad that [7] concluded that there is no
segmented Image can be selected on the basis of visual segmentation method that is better than the other in all
inspection and the value of the parameters for that domains. We believe that with the use of universal
segmented image can be chosen as a result.
evaluation methods we can be able to find the
Similarly, all the algorithms are applied on an image and
for every segmentation algorithm; the result is displayed segmentation techniques that we may say are better than
on same figure window. others in all domains.

V. PERFORMANCE EVALUATION VI. CONCLUSION

 
After analysis of various image segmentation algorithms
There have been many image segmentation methods
and the comparison of the results of each algorithm
created and being created using many distinct approaches separately with different parameter's value using
and algorithms but still it is very difficult to assess and MATLAB, the conclusion is that:
compare the performance of these segmentation In Edge Based Segmentation Algorithms, the Canny
techniques [5]. Researchers would evaluate their image Algorithm produced the best segmentation in comparison
segmentation techniques by using one or more of the of Sobel, Prewitt and LoG.
following evaluation methods in fig.8. In Thresholding Based Algorithms, the Adaptive
Thresholding and Otsu Thresholding produced good
results. The Adaptive thresholding produced the good
edges and Otsu Thresholding recognized the object very
well. In Region Based Algorithms, the split and merge
method produced the better result.
In Clustering Based Segmentation, the mean shift
method produced the good result. When the K-means and
Fuzzy-C means methods are compared, the Fuzzy C-
means is better than the K-means method.
In Graph Based Algorithms, Normalized-cut is used to
cut an image into specified number of cuts. Other methods
are interactive methods. In N-cut, the foreground and
background area is selected by a user.
To produce a good result with a single technique for the
images of all the applications, the further research is
required and from the proposed system it can be concluded
that the further research should concentrate on such
Fig.8 an Overview of Evaluation Techniques techniques in which the user's interaction is involved so
that the segmented result can be improved after automatic
The full description of the above evaluation methods segmentation.
can be found from [6]. Most of these methods ideally
should be domain independent but in reality they are VII. REFERENCES
domain dependent. It is believed that it is difficult to [1] R. C. Gonzalez, R. E. Woods, "Digital Image Processing," 2nd
develop a single model that applies to all image objects . Prentice-Hall Inc, 2002.
[2] L. G. Shapiro and G. C. Stockman, "Computer Vision," Prentice-
Both the subjective and objective evaluation have been Hall Inc., New Jersey, 2001, pp. 279-325.
used to evaluate segmentation techniques but within a [3] http://en.wikipedia.org/wiki/MATLAB.
domain dependent environment [6]. It can be appreciated [4] http://www.math.utah.edu/~eyre/computing/matlab-intro/.
[5] ZHANG, H., FRITTS, J. E. & GOLDMAN, S. A. 2008. Image
that whatever method used in a specific domain has been Segmentation Evaluation: A survey of unsupervised methods.
used to compare the segmentation technique in that Computer Vision and Image Understanding, 10(2):260 280.
[6] POLAK, M., ZHANG, H. & PI, M. 2009. An evaluation metric
domain. These methods have been used to adjust for image segmentation of multiple objects. Image and Vision
parameters of the segmentation techniques in order to Computing, 27(8):1223-1227.
[7] HU, S., HOFFMAN, E. A. & REINHARDT, J. M. 2001.
solve the following problems in segmentation area: Automatic Lung Segmentation for Accurate Quantization of
Volumetric X-Ray CT Images. IEEE, 20(6):490-498.
• The segmented region might be smaller or larger than [8] ZHANG, Y. J. 2001. A review of recentevaluation methods for
the actual image segmentation. Paper presented at the International
Symposium on Signal Processing and its Applications (ISSPA),
• The edges of the segmented region might not be Kuala Lumpur.
connected [9] UDUPA, J. K., LEBLANC, V. R., ZHUGE, Y., IMIELINSKA,
• Over or under-segmentation of the image (arising of C., SCHMIDT, H., CURRIE, L. M., et al. 2006. A framework
pseudo edges or missing edges). for evaluating image segmentation algorithms. Computerized
Medical Imaging and Graphics, 30( ):75-87.

Copyright © 2012 IJEIR, All right reserved 54

 International Journal of Engineering Innovation & Research
Volume 1, Issue 1, ISSN 2277 – 5668
 

[10] VARSHNEY, S. S., RAJPAL, N. & PURWAR, R. 2009.

Comparative study of image segmentation techniques and object
matching using segmentation Paper presented at the International
Conference on Methods and Models in Computer Science.
[11] WANG, L., HE, L., MISHRA, A. & LI, C. 2009. Active
contours driven by local Gaussian distribution fitting energy.
Signal Processing.
[12] WANG, Y., GUO, Q. & ZHU, Y. 2007. Medical image
segmentation based on deformable models and its applications
Springer:2
[13] BOUCHERON, L. E., HARVEY, N. R. & MANJUNATH, B.
S. 2007. A quantitative object-level metric for segmentation
performance and its application to cell nuclei. springer-Verlag
2007:208-219.

AUTHOR’S PROFILE

Gajendra Singh Chandel received his Bachelor of

Engineering degree in Information Technology from
Oriental Institute of Science and Technology,RGPV
Bhopal In 2007, M.P., India. He has completed his
M.Tech (Master of Technology) degree in from
Computer Science & Engineering Lakshmi Narayan
College of Technology, RGPV Bhopal, In 2010 M.P.,
ndia. Presently he is HOD Of Computer Science
Engineering & Information Technology Department in
SSSIST, Sehore M.P., India. He is having 4 Yrs of
teaching experience . He has published 12 papers in
refereed International/National Journal and Conference
including IEEE. He is a Member of Easy Chair
Conference System.
                                                                                                  
Ravindra Kumar Gupta received his M.Tech (Master
of Technology) degree in Computer Science &
Engineering from Sri Satya Sai Institute Of Science &
Technology, RGPV Bhopal, M.P., India. Presently he is
Assistant Professor Of Computer Science & Engineering
Department in SSSIST; Sehore M.P. India. He is having
Yrs of teaching experience .He has published 29papers
in referred International/National Journal & conference.
He is a Member of Easy Chair Conference System.

Deepika Khare pursuing her M. Tech. (Master of

Technology) degree in Information Technology from Sri
Satya Sai Institute Of Science & Technology, RGPV
Bhopal, M.P., India. She has published 4 papers in
referred International/National Journal & Conferences.

Sumita Verma    pursuing her M. Tech. (Master of

Technology) degree in Computer Science & Engg. from
Sri Satya Sai Institute Of Science & Technology, RGPV
Bhopal, M.P., India. She has published 4 papers in
referred International/National Journal & Conferences.

Copyright © 2012 IJEIR, All right reserved 55