ISSN 2394-3777 (Print)

ISSN 2394-3785 (Online)

Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 4, Special Issue 21, August 2017

MODERN GRAPH-THEORETIC APPROACH FOR

CLUSTERING AND IMAGE SEGMENTATION
K. Venkatasubramanian Dr. S.K.Srivatsa
Research Scholar, SCSVMV University, Kanchipuram Retd. Prof. Department of CSE,
Associate Professor, Ramachandra College of Engineering, Eluru Anna University ,Chennai - 602 025.
email: [email protected] email: [email protected]
Dr. C. Parthasarathy
Assistant Professor, Dept. of IT,
SCSVMV University, Kanchipuram
email:[email protected]

ABSTRACT
Image segmentation is an interesting and developing concept in the field of computer vision. The application of different graph theoretic
technique became more useful in image segmentation process. In this paper a dense region method of clustering is used for image segmentation. Here
the concept of dense region algorithm is used to find out the maximum dense region of the image by applying gaussian filter and some morphological
operators. By using the constraint retrieved images that are located close to each other in the feature space, Image retrieval is done. This approach is
experimented with real world images which show the effectiveness of the method.

Keywords: Clustering, Images, Maximum, Segmentation

I.INTRODUCTION Clustering is a process by which we can group together the

objects such that the objects belongs to the same cluster will
Image segmentation is very important and challenging
have same property but objects belongs to different cluster will
problem and a necessary role in image analysis as well as in
have different property. There are mainly two types of
high-level image interpretation and understanding such as robot
clustering namely partitioning and hierarchical. At present
vision, object recognition, and medical imaging. The goal of
other technique has been developed. Many of them are hybrid
image segmentation is to partition an image into a set of
in nature. Nevertheless, based on the basic architecture,
disjoint regions with uniform and homogeneous attributes such
clustering can be classified as density-based, grid based, model-
as intensity, color, tone or texture, etc. Many different
based, sample-based etc. Density-based clustering methods are
segmentation techniques have been
done based on density. It is believed that density-based
developed till now. In this paper, a clustering based method for
clustering methods have the potential to reveal the structure of
image segmentation will be considered. Classification can be of
a spatial data set in which different point processes overlap.
two types i.e. supervised and unsupervised. Unsupervised
classification is known as clustering. In supervised Clustering techniques are a standout amongst the most
classification we need some prior information about the utilized calculations as a part of Image segmentation, uniquely
classification. But in the unsupervised classification no prior used to think about a new proposition with them. There are
information about the classification is needed. i.e. It basically two types of method of clustering in graph theoretic
automatically generates the clusters. approach 1) Hierarchical clustering 2) Partitional clustering 299
 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 4, Special Issue 21, August 2017

algorithm. Where the primary contrast between them for its optimization. We apply the method to cluster, digital
is that Hierarchical strategies create a settled arrangement of color photographs and present results to demonstrate its
partitions, while partitional strategies deliver stand out usefulness and effectiveness[1].Yixin Chen et.al proposed a
allotment. Even though Hierarchical strategies can be more new technique, CLUster-based retrieval of images by
exact, partitional strategies are utilized as a part of utilizations unsupervisedlearning (CLUE), for improving user interaction
including extensive information sets, similar to the ones related with image retrieval systems by fully exploitingthe similarity
with pictures, because of the reality the development of a tree information. CLUE retrieves image clusters by applying a
structure is computationally restrictive. The complexity of graph-theoretic clustering algorithm to a collection of images in
Hierarchical algorithm is O(n3d) where n is the number of the vicinity of the query. Clustering in CLUE is dynamic. In
pixels of the image and d is the number of features. Whereas particular, clusters formed depend on which images are
2
the complexity of partitional algorithm is O(n d). retrieved in response to the query. CLUE can be combined with
any real-valued symmetric similar measure (metric or
The basic issue for picture division inspirations
nonmetric). Thus, it may be embedded in many current CBIR
driving most partitional bunching calculation is that they don't
systems, including relevance feedback systems. The
use spatial information , alternately in the methodologies which
performance of an experimental image retrieval system using
utilizes the area use themselves, such information (in truth
CLUE is evaluated on a database of around 60, 000 images
spatial information is the reason of these procedures).
from COREL. Empirical results demonstrate improved
Something else, bunching calculation acknowledges that pixels
performance compared with a CBIR system using the same
having a spot with the same gathering have a tantamount
image similarity measure. In addition, results in images
behavior, which is illuminated by uniform components like
returned by Google’s Image Search reveal the potential of
faint levels, surface or shading. Checking the correlative way of
applying CLUE to real world image data and integrating CLUE
grouping and district systems, our approach bets for the
as a part of the interface for keyword-based image retrieval
coordination of both procedures with a particular final objective
systems[2].
to handle the issues that both strategies bear when are used
MuthuKumaravel proposed an approach in which a
freely. In composing there are assorted suggestions for this
novel image retrieval scheme CLUster-based retrieval of
joining. For example, one of the first works organizing both
images by unsupervised learning which tackles the semantic
systems is that in which spatial goals are fused by the usage of
gap problem based on a hypothesis: semantically images tend
a Gibbs Random Field, which hunt down neighborhood power
to be clustered in some feature space. CLUE attempts to
assortments. Of course, the essential plan contains on including
capture semantic concepts by learning the way that images of
the pixels encourages as artificial components in the bunching
the same semantics are similar and retrieving image clusters
procedure. In this paper an experimental study of density-based
instead of a set of ordered images. Clustering in CLUE is
image clustering is done.
dynamic. In particular, clusters formed depend on which

II. LITERATURE REVIEW images are retrieved in response to the query therefore; clusters
give the algorithm as well as the user’s semantic relevant clues
GuopingQiu proposed a method which simultaneously as to where to navigate. CLUE is a general approach that can be
models and clusters, large sets of images and their low-level combined with any real-valued symmetric similarity measure.
visual features. A computational energy function suited for co- Thus, it may be embedded in many current CBIR systems.
clustering images and their features is first constructed and a
Hopfield model based stochastic algorithm is then developed
 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 4, Special Issue 21, August 2017

Experimental results based on a database of about 60,000 load the image.The image is loaded by using the image data
images from COREL demonstrate improved performance. function.Then the image is shown by using
ArpanGarai et.al proposed an approach in which the imshowfunction.After the image is loaded, the next step is to
density based fuzzy c means (DBFCM) clustering is presented. create a gaussianfilter.In this we are creating Gaussian-Filter to
It is divided into two steps, first different dense region is found remove the noise which is present in the image.The Gaussian-
using kth nearest neighbour then fuzzy c means segmentation is Filter is helpful to get the very clear and predefined image. The
done on each dense region. The DBFCM is implemented upon FSPECIAL function is used to create the predefined 2-D filter
some of the satellite images to get the segments as experimental and it also creates the parameters for it.
results. Then it is compared with the conventional fuzzy c
After the Gaussian-Filter is added to the image, then
means approach[4].
the morphological operation is performed.The morphological
operation is done by using STREL function.It creates a disk-
Nikhil R Pal et.al proposed an approach in which various
shaped structuring element of the given input image. Based on
attempts have been made to cover both fuzzy and non-fuzzy
the closed image, the maximum dense region is found by using
techniques including color image segmentation and neural
the IMREGIONALMAX and IMDILATE function. The image
network based approaches. Adequate attention is paid to
returned from IMREGIONALMAX doesn't always have just
segmentation of range images and magnetic resonance images.
single pixels set to 1 (to indicate a maximum). The output
It also addresses the issue of quantitative evaluation of
image often contains clusters of pixels because the neighboring
segmentation results[5].
pixels in the input image can have equal values, and are
III.EXISTING SYSTEM:
therefore both counted as maxima. To avoid this IMDILATE
In the Existing System, the Image Grouping and Retrieval is function is used. To dilate these points with IMDILATE just to
done by using the Minimum Spanning Tree andModified make them easier to observe the image, which makes an even
hierarchical Clustering Algorithm. One of the drawbacks of bigger cluster of pixels centered on the maximum region in an
hierarchical clustering algorithm is that the time complexity is image.
high, generally it’s in the order of O(n2logn), n being the V.ALGORITHM FOR DENSE-REGION:
number of data points.Another setback in the hierarchical The following are the steps to be followed in Dense-Region
clustering algorithm is that it is not easy to define levels of Algorithm.
clusters. Only the images which are close to the feature space, Step 1.Label all points as core, border, or
image retrieval is done. The image retrievel which is based on noise points.
finding the maximum dense region in an image has not been Step 2.Eliminate noise points
addressed and implemented. Step 3.Adding the Gaussian-Filter to get the
predefined Image.
IV.PROPOSED SYSTEM:
Step4.Perform the Morphological
In the proposed system, we are going to address the Operation
drawbacks which are faced in the existing system. (i.e) The Step 5.Find the Region where local
image retrieval which is based on finding the maximum dense Maxima is occurred
region for an image has been found out by using dense region Step 6.Getting the maximum dense region
algorithm. We are going to find the maximum dense region for present in the image.
the lattice image by using the following steps.TheFirst step is to
 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 4, Special Issue 21, August 2017

VI. EXPERIMENTS AND RESULTS Figure 3 Shows the Closed Image

The experiment performed using matlab to assess the The Figure4 finally shows a maximum dense region in a lattice

performance of dense-region clustering algorithm. image.The maximum dense region image is created by using
IMREGIONALMAX function. [3] proposed a system in which
The Figure1 shows the lattice image for which the maximum OWT extracts wavelet features which give a good separation of
dense region is going befindout. different patterns. Moreover the proposed algorithm uses
morphological operators for effective segmentation. From the
qualitative and quantitative results, it is concluded that our
proposed method has improved segmentation quality and it is
reliable, fast and can be used with reduced computational
complexity than direct applications of Histogram Clustering.
The main advantage of this method is the use of single
parameter and also very faster. While comparing with five
color spaces, segmentation scheme produces results noticeably
better in RGB color space compared to all other color spaces.
Figure 1 Shows the Original Lattice Image

The Figure2 shows the Gaussian-Filter Image.After adding the

Gaussian-Filter the predefined image has been created.

Figure 4 Shows the Maximum Dense Region in an Image

VII. CONCLUSION:
Figure 2 Shows the Gaussian-Filter Image These methods of proposed algorithm are discussed the
The Figure3 shows the Closed Image.The closed image is representative methods of dense region clustering. In this
created by using STREL function in morphological operation. paper, segmentation an image into displace areas, such that
each region satisfies a certain predefined partition criterion. In
this paper, the problem of finding the maximum dense region in
an image has been resolved by applying the Gaussian-Filter.
The proposed approach improves accuracy rate compared to
existed methods which are used to retrieve the maximumdense
region for an Image.
 ISSN 2394-3777 (Print)
ISSN 2394-3785 (Online)
Available online at www.ijartet.com

International Journal of Advanced Research Trends in Engineering and Technology (IJARTET)

Vol. 4, Special Issue 21, August 2017

VIII. REFERENCES:

[1] Yixin Chen, James Z. Wang and Robert Krovetz,

“CLUE: Cluster-Based Retrieval Of Images By
Unsupervised Learning, July 23, 2004.
[2] Arpan Garai1, Kalyani Mali2, “Density Based Fuzzy
C Means (DBFCM) Image Segmentation”, Vol. 2,
Issue 8, August 2013.
[3] Christo Ananth, A.S.Senthilkani, Praghash.K, Chakka
Raja.M., Jerrin John, I.Annadurai, “Overlap Wavelet
Transform for Image Segmentation”, International
Journal of Electronics Communication and Computer
Technology (IJECCT), Volume 4, Issue 3 (May 2014),
pp-656-658
[4] Dr. A. MuthuKumaravel, “Cluster Based Image
Retrieval”, Vol. 1, Issue 5, July 2013.
[5] V. V. Raghavan and C. T. Yu. A comparison of the
stability characteristics of some graph theoretic
clustering methods. IEEE Trans. Pattern Anal.
Machine Intell., 3:393–402,1981.
[6] A. K. Jain and R. C. Dubes. Algorithms for Clustering
Data.Prentice Hall, Englewood Cliffs, NJ, 1988
[7] Nikhil R Pal, Sankar K Pal, “A review on Image
Segmentation Techniques”, Volume 26, Issue 9.
[8] MassimilianoPavan and Marcello Pelillo, “A New
Graph-Theoretic Approach to Clustering and
Segmentation, 2003
[9] Satu Elisa Schaeffer, “Graph Clustering”, 2007.
[10] R. A. Fisher. The Genetical Theory of Natural
Selection.Oxford University Press, London, UK, 1930.
[11] D. Forsyth and J. Ponce. Computer Vision: A Modern
Approach.Prentice-Hall, Englewood Cliffs, NJ, 2002.
[12] M. Pavan, M. Pelillo, and E. Jabara. On the
combinatorics of standard quadratic optimization.
Forthcoming, 2003.

303