A comparative study on classification of image segmentation methods with a focus on graph based techniques

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
__________________________________________________________________________________________
Volume: 03 Special Issue: 03 | May-2014 | NCRIET-2014, Available @ http://www.ijret.org 310
A COMPARATIVE STUDY ON CLASSIFICATION OF IMAGE
SEGMENTATION METHODS WITH A FOCUS ON GRAPH BASED
TECHNIQUES
Basavaprasad B1
, Ravi M2
1
Asst. Professor, Department of Computer Science, Govt. First Grade College, Raichur, Karnataka, India
2
Asst. Professor, Department of Computer Science, Govt. First Grade College, Manvi, Raichur, Karnataka, India
Abstract
Image segmentation is the most precarious functions in image processing and analysis. Basically segmentation consequences
influence all the subsequent processes of image analysis such as object description and illustration, characteristic dimension,
and even the subsequent higher level tasks such as classification of object. Hence, image segmentation is the most important and
critical process for assisting the, depiction, delineation and visualization of regions of interest in any image. Physical segmentation of
an image is not only a tiresome and time consuming process, but also not exceptionally accurate particularly with the increasing
imaging modalities and uncontrollable quantity of images that need to be observed. Therefore it becomes essential to examination
current methodologies of image segmentation using computerized algorithms that are precise and entail as little user
interaction as probable in particular for medical images. In the image segmentation process, the anatomical organization or the
region of interest needs to be defined and extracted out so that it can be viewed independently. In this comparative paper we venture
the significant place of segmentation of images in pulling out information for decision making. Afterwards we present the most
relative and general methods which we have classified into three categories, first category represents pixel based, second edge based
and third represents region based methods in particular graph based methods, highlighting the weaknesses and strengths according to
appropriateness for image segmentation applications.
Keywords: Classification, Segmentation, Techniques, Graph based, Pixel-based, Edge-based, Histogram, Thresholding,
Clusters, Comparative.
-----------------------------------------------------------------------***----------------------------------------------------------------------
1. INTRODUCTION
Generally, image segmentation can be defined as a
procedure for partitioning a digital image into multiple
regions (set of pixels). Image segmentation separates an
image into regions with no overlapping. Each region
possesses different features including color, intensity, texture
or other statistical properties. The main aim of image
segmentation is to simplify and transform the representation of
an image into something that is further meaningful and
simpler to analyze. The result of image segmentation is either
a set of regions that together cover up the entire image or a set
of contours extracted from the image [1]. From the
foregoing general definition, a more specific definition can be
given; the process of delineating and separating the
anatomical structure (region of interest) so that it can be
viewed individually in order to achieve important objectives
such as patient diagnosis, radiotherapy planning, and
advanced surgical planning and research. Therefore, TRUS is
more popular for screening and biopsy guidance for
prostate cancer. It is a safe means of imaging the human
body because it emits no ionizing radiation, and it uses sound
waves of low-power. It is comparatively contemptible and
transportable to handle. If the domain of the image is
known by I, next, the image segmentation problem is to
determine the sets whose union is the complete image I.
Therefore, the sets that build up image segmentation must
satisfy:
……………………….. (1)
Where
Where and each is a connected
region. Ideally, segmentation method defines those sets that
belong to distinctive anatomical structures or regions of
interest in the image [4]. One direct way to segment an
image is by applying an edge detection technique,
especially If the image consists of clear objects on a
different intensity background. However, this method can fail,
if the image contains noise. It is usually necessary to conduct
validation experiments so as to quantify the performance of a
segmentation technique. It can be done by comparing the
automated segmentation with the manual segmentation of an

__________________________________________________________________________________________
expert. An alternative is using physical phantoms or
computational Phantoms. There are numerous image
segmentation techniques. However, there is no single or
standard segmentation technique that works well for all
images. In this paper we have presented a systematic study on
image segmentation techniques which we classified into three
categories. Firstly pixel-based methods which are explained in
section 2, secondly edge-based methods presented in section 3
and finally graph-based methods are presented in Section 4.
Fig 1: Image Segmentation Techniques
The division of an image into meaningful structures, image
segmentation, is often an essential step in object
representation, image analysis, apparition, and many other
image processing tasks. We focused on how to analyze and
correspond to an object, except our assumption the sets of
pixels that identified that object was known in advance. In this
section, we will focus on methods that find the particular
pixels that make up an object. A great variety of segmentation
methods has been proposed in the past decades, and some
categorization is necessary to present the methods properly
here. A disjunction categorization does not seem to be
possible though, because even two very different segmentation
approaches may share properties that defy singular
categorization1. The categorization presented in this chapter is
therefore rather a categorization regarding the emphasis of an
approach than a strict division. The following categories are
used:
• Pixel based segmentation is the process of partitioning
a digital image into multiple segments or groups of pixels, also
known as super pixels.
• Threshold based segmentation, Histogram thresholding
and slicing techniques are used to segment the image. They
may be applied on an image directly, but also can be
combined with preprocessing and post-processing techniques.
• Edge based segmentation. With this method, the edges
which are detected in an image are assumed to represent
object boundaries, and used to identify these objects.
• Region based segmentation. Where an edge based
technique may attempt to find the object boundaries and then
locate the object itself by fulfilling them in, a region based
method takes the opposite approach Structural Segmentation
Methods
These methods utilize information about the structural features
of the image to implement segmentation of the target image.
Some of the common structural methods include edge-
detection, graph searching, deformable models, is on
surface and level set. We highlight some.
2. PIXEL BASED IMAGE SEGMENTATION
2.1 Thresholding Methods:
We have taken thresholding method as an example to explain
pixel based image segmentation. The thresholding technique is
the simplest image segmentation technique. It is based on
the assumption that the objects and the background in the
image have a bimodal distribution. Typically, these
assumptions not valid for most images, especially medical
ones. The key point in this segmentation technique is to
determine an intensity rate, which is called the threshold,
which separates the desired classes. The main advantages of
deformable models are their ability to directly generate
closed parametric curves or surfaces from images and their
incorporation of a smoothness constraint that provides
robustness to noise and spurious edges. Moreover, they offer
sub-pixel accuracy for the boundary representation, which
may be important to medical applications. An important
drawback of the snakes is that they require manual
interaction to place an initial model or to choose
appropriate initial parameters. Besides, they cannot be used
in non-interactive applications, unless they are initialized close
to the structure to be segmented because they are designed as
interactive models. Thresholding segmentation transforms
input image, I, to a binary image, g, by grouping the
pixels with intensities, higher than the threshold into one the
class, and the further pixels into another class. For any 2D
digital image [16],
and
………………….. (2)
The technique of thresholding is defined as:
………………...(3)
Some images can be segmented by using more than one point
of thresholding, also called as multi-thresholding, as, where
and are the two thresholds segmenting the image. The
output image, resulting from multi-thresholding, is no longer
binary, but consists of a limited number of grey levels as

__________________________________________________________________________________________
follows:
… (4)
is the image grey levels number in the image (256).
Fig 2: Histogram showing three apparent classes using two
thresholds and .
3. EDGE BASED IMAGE SEGMENTATION
The problem of detecting edges in an image is a basic function
in analysis of images. It engages characterizing the object
boundaries in an image using edge information. It then makes
edge detection very handy for segmentation. Edge detection
techniques are commonly used for detecting edges in an image
to perform segmentation process. The traditional edge
detection methods usually use the edge detection operators,
which are based mainly on the gradation such as Sobel,
Robert, and Prewitt edge detectors . Usually edges occur at
the point of intersection of two regions with intensities which
are varying. The advantage of these techniques is that they
work very well only on images with high-quality disparity
between different regions. Their disadvantages include; they
detect all the edges; hence, it is very difficult to find the
relation between the edges and the region of interest. In
addition, those algorithms are sensitive to noise. Edge-based
segmentation corresponds to a huge group of methods based
on information concerning to edges in an image.
 Edge-based segmentations depend on edges found in an
image by the edge detecting operators. These edges spot
image locations of discontinuities in color, texture, gray level
etc.
 Image resulting from the method of edge detection cannot
be used as a result of segmentation. Accompanying processing
steps must go after to merge edges into edge chains that
correspond to improve with borders in the image.
 The final desire is to reach at least a fractional
segmentation; that is, to bunching neighboring edges into an
image where only edge chains with an association to existing
things or image parts are in attendance.
 The more aforementioned information that is available to
the segmentation procedure, the better the image segmentation
results that can be attained.
 The most general problems of edge-based image
segmentation are an edge occurrence in locations anywhere
with no border and no edge exists where an actual border is
there.

Fig 3: Edge based image segmentation flow.
4 REGION BASED METHODS
4.1 Region Growing Methods
We have taken region growing method as an example to
explain region based image segmentation. Region growing
takes advantage of an important fact that the pixels close
together have similar gray values. The main idea of this
approach consists in the observation that the pixels belonging
to one element of the object can possess similar properties, for
example, the gray value. Therefore if the considered pixel has
gray value that is near the common gray value of the region,
this pixel can be associated into this region. This technique
generally consists of the following three steps; Initially, a new
region consists of only one pixel which is called the seed pixel
is selected. This pixel is determined by user, and issued as a-
priori information. The given initial pixel determines the8-
neighbour where all pixels are sequentially checked. If a pixel
does not belong to any other regions it can be associated into
the given region if the difference between the gray value of the
pixel and the averaged gray value of the whole region is less
than a predefined threshold, T. Second, when the pixel is
connected to the region the averaged gray value of this region
is re-calculated. After all the neighboring pixels are reviewed,
new neighborhood for the enlarged region will be determined
and the first step is.
4.2 Graph Search Algorithm:
Graphs are an abstract representation consisting of a set of
vertices and a set of
edges where
indicates that there is an edge (arc) from vertex i to vertex

__________________________________________________________________________________________
j. Each edge has an associated weight representing the cost of
transition from , as shown in Fig. In these algorithms,
edges and surfaces are represented as graphs, and the
algorithm tries to find the lowest-cost path between two
nodes of the graph using any search algorithm. Those
algorithms have the advantage that they can perform well
even if the separations between regions are broken. Their
disadvantage is that of requiring surfaces to be represented as
graphs, which could be tricky [1]. These graph based
algorithms are playing a vital role in image segmentation
especially in medical image processing which is topic
emerging as most active research from the past two decades
[8].
Fig 4: Both Connected and disconnected graphs.
5. RESULTS
The result of this is paper is mentioned in the below images
and table.
a) Original Cameramen Image
b) Pixel Based Image Segmentation
c)Edge Based Image Segmentation
d) Graph Based Image Segmentation
Fig 5: Comparative Segmented Images
Table 1: Comparison of classified methods
Method Advantage
Segmentation
effect
Pixel Based
Segmentation
More Complexity Normal
Edge Based
Segmentation
Average
Complexity
Average
Graph Based
Segmentation
Very low
complexity
Good
6. CONCLUSIONS
We have thoroughly discussed the useful methods of pixel
based, edge based and region based (graph) methods of image
segmentation. This systematic comparison study is helpful for
individual researchers to do research in the ground of image
segmentation. These Image segmentation techniques are
extremely efficient particularly, graph based image
segmentation methods which comes under the third category
i.e., region based methods. Medical image processing is one of
the main dynamic research topics in image processing [18].
Most modern research in image segmentation has tinted the
potential of graph based techniques for medical applications.
In order to implement graph theory in image segmentation
professionally in particular in medical image processing we
need to set up implementation between mathematical
outstanding junior scientists and biological scientists, and

__________________________________________________________________________________________
describe the sketch to build up the new tools on this domain.
The motivation should be on the study of properties of Euler
graphs, minimal spanning trees, Fuzzy graphs, shortest paths
trees, minimal cuts and Normalized cuts and we re-examine
these ideas for image segmentation purposes. From this
comparison study we conclude that there is no universal
segmentation method that can be executed for all types of
images, but a number of techniques do healthier than others
for particular types of images representative better
performance can be attained by selecting appropriate
algorithm or combination of suitable techniques [29].
REFERENCES
[1]. Basavaprasad B. and Ravindra S. Hegadi, “A Survey on
Traditional and Graph Theoretical Techniques for Image
Segmentation", Recent Advances in Information Technology,
International Journal of Computer Applications, Pages: 38-46,
2014.
[2]. Shanbehzadeh J. and Rastgarpour M., “Application of AI
Techniques in Medical Image Segmentation and Novel
Categorization of Available Methods and Tools”,
International Multi-Conference of Engineers and Computer
Scientists Proceedings, Volume: 1, Hong Kong, 2011.
[3]. Zhang, Y. J, “An Overview of Image and Video
Segmentation in the last 40 years”, 6th International
Symposium on Signal Processing and Its Applications
Proceedings, Pages: 144-151, 2001.
[4]. Wahba Marian, “Automated Modified Region Growing
Technique for Prostate Segmentation in Trans-Rectal
Ultrasound Images”, Master’s Thesis, Department of
Electrical and Computer Engineering, University of Waterloo,
Waterloo, Ontario, Canada, 2008.
[5]. Shapiro L. G. and Stockman G. C., Computer Vision,
Prentice-Hall Publications, New Jersey, 2001.
[6]. Pham D. L, Prince J. L. and Xu C, “Current Methods in
Medical Image Segmentation”, Biomedical Engineering,
Annual Review, Volume: 2, Pages: 315-38, 2000.
[7]. Lakare S., and Kaufman A., “3D Segmentation
Techniques for Medical Volumes”, A Center for Visual
Computing”, Department of Computer Science, State
University of New York, December 2000.
[8]. Viergever M. A and Lobregt S., “A Discrete Dynamic
Contour Model”, Medical Imaging, an IEEE Transactions,
Volume: 14, Pages: 12–24, 1995.
[9]. Kass M., Terzopoulos D. and Witkin A., “Snakes: Active
Contour Models”, IJCV, Volume: 1, Pages: 321–331, 1988.
[10]. Szeliski R. and Terzopoulos D., “Tracking with Kalman
Snakes”, Active Vision and Artificial Intelligence, Pages: 3–
20, Cambridge, the MIT Press, 1992.
[11]. Hamarneh G., “Towards Intelligent Deformable Models
for Medical Image Analysis”, Ph.D. Thesis, Department
signals and Systems, Electrical and Computer Engineering
School, Technology of Chalmers University, 2001.
[12]. Awad, J.G.E., “Prostate Segmentation and Regions of
Interest Detection in Transrectal Images of Ultrasound”,
Thesis of Doctor of Philosophy, University of Waterloo,
Waterloo, Ontarion, Canada, 2007.
[13]. Sahoo P. K., Farag A. A., and Yeap Y. P., Threshold
Selection Based on Histogram Modeling, IEEE International
Conference on Systems, Man and Cybernetics, Volume: 1,
Pages: 351-356, 1992.
[14]. Pathak S. D., Grimm P. D., Chalana V., and Kim
Y.,Pubic Arch Detection in Transrectal Ultrasound Guided
Prostate Cancer Therapy, IEEE Transactions on Medical
Imaging, Volume: 17, Pages: 762-771, 1998.
[15]. Potocnik B., Solina F., and Zazula D., “Classical Image
Processing vs. Computer Vision Techniques in Computer-
Assisted and Automated Detection of Follicles in Ultrasound
Images of Ovary”, Image Processing and its Applications, 6th
International Conference, Volume: 2, Pages: 551-555, 1997.
[16]. Lee C., Huh S., Ketter T. A., and Unser M.,
”Unsupervised Connectivity-Based Thresholding
Segmentation of Midsagittal Brain MR Images, Computers in
Biology and Medicine, Volume: 28, Pages: 309-338, 1998.
[17]. Schalkoff R.J., Pattern Recognition: “Statistical,
Structural and Neural Methods”, John Wiley and Sons, 1992.
[18]. Pham D. L., Prince J. L. and Xu C., “Survey of present
Methods in Medical Image Segmentation”, Annual Review of
Engineering (Biomedical), 1998.
[19]. Xu R., and Wunsch D., “A Survey of Clustering
Algorithms”, Neural Networks IEEE Transaction, Volume:
16, May 2005.
[20]. Hebert T. J., “Quick Iterative Segmentation of High
Resolution Medical Images”, Nuclear Science, IEEE
Transaction, Volume: 44, Pages: 1363–1367, 1997.
[21]. Hall L.O., Bensaid A.M., Bezdek J.C. Clarke L.P.,
Velthuizen R.P.Silbiger M.S., “A Comparison of Neural
Network and Fuzzy Clustering Techniques in Segmenting
Magnetic Resonance Images of the Brain”, IEEE
Transactions on Neural Networks, Volume: 3, Pages: 672–
682, 1992.
[22]. Gelenbe E., Krishnan K.R.R. and Feng Y., “A Neural
Network Methods for Volumetric Magnetic Resonance
Imaging of the Human Brain”, IEEE Proceedings, Volume:
84, Pages: 1488–1496, 1996.
[23]. Hall L.O., Clarke L.P. and Bezdek J.C., “Review of MR
Image Segmentation Techniques Using Pattern Recognition”,
Medical Phys, Volume: 20, Pages: 1033–1048, 1993.
[24]. Reddick W.E., Cook E.N., Glass J.O., Elkin T.D., and
Deaton R.J., “An Automated Segmentation and Classification
of Multispectral Magnetic Resonance Images of Brain Using
Artificial Neural Networks”, IEEE Transactions on Medical
Imaging, Volume: 16, Pages: 911–918, 1997.
[25]. Vilarino D.L., Cabello D., Brea V.M. and Pardo J.M.,
“Discrete-Time CNN for Image Segmentation by Active
Contours”, Proceedings of Pattern Recognition, Volume: 19,
Pages: 721–734, 1998.
[26]. Tournoux P and Talairach J. “Co-Planar Stereotaxic
Atlas of the Human Brain”, Proportional System on 3D: An
Approach to Cerebral Imaging, Thieme Medical Publisher,
1988.

__________________________________________________________________________________________
[27]. Sandor S., and Leahy R., “Surface-Based Labeling of
Cortical Anatomy Using a Deformable Atlas”, Medical
Imaging, IEEE Transactions, Volume: 16, Pages: 41–54,
1997.
[28]. Davatzikos C., Vaillant, M., Resnick S., Prince J. L.,
Letovsky S and Bryan R. N., “A Computerized Method for
Morphological Analysis of the Corpus Callosum”, An
International Journal of Computer Application, Volume: 20,
Pages: 88–97, 1996.
[29]. Basavaprasad B. and Ravindra S. Hegadi, “An Improved
GrabCut Technique for segmentation of Color Images”,
Recent Advances in Information Technology, an International
Journal of Computer Applications, Pages: 5-8, 2014.
[30]. Garcia-Consuegra J., Cisneros G. and Navarro E., “A
Sequential Echo Algorithm Based on the Integration of
Clustering and Region Growing methods”, Proceedings of
IGARSS, Volume: 2, Pages: 648-650, 2000.
[31]. Pan W., “Application of Region Growing Algorithm on
Handwritten Numerical Recognition”, thesis for Master of
Science, Department of Computer Science and Engineering,
Tatung University, 2004.

A comparative study on classification of image segmentation methods with a focus on graph based techniques

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to A comparative study on classification of image segmentation methods with a focus on graph based techniques (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

A comparative study on classification of image segmentation methods with a focus on graph based techniques