An Unsupervised Cluster-based Image Retrieval Algorithm using Relevance Feedback

International Journal of Managing Information Technology (IJMIT) Vol.3, No.2, May 2011
DOI : 10.5121/ijmit.2011.3202 9
An Unsupervised Cluster-based Image Retrieval
Algorithm using Relevance Feedback
Jayant Mishra1
, Anubhav Sharma2
and Kapil Chaturvedi3
1
Department of Computer Application, UIT, RGPV, Bhopal
rgtu.jayant@gmail.com
2
Department of Computer Science, RITS, Bhopal
ac.anubhav@gmail.com
1
Department of Computer Application, UIT, RGPV, Bhopal
kplchaturvedi@yahoo.com
ABSTRACT
Content-based image retrieval (CBIR) systems utilize low level query image feature as identifying similarity
between a query image and the image database. Image contents are plays significant role for image
retrieval. There are three fundamental bases for content-based image retrieval, i.e. visual feature
extraction, multidimensional indexing, and retrieval system design. Each image has three contents such as:
color, texture and shape features. Color and texture both plays important image visual features used in
Content-Based Image Retrieval to improve results. Color histogram and texture features have potential to
retrieve similar images on the basis of their properties. As the feature extracted from a query is low level, it
is extremely difficult for user to provide an appropriate example in based query. To overcome these
problems and reach higher accuracy in CBIR system, providing user with relevance feedback is famous for
provide promising solution.
KEYWORDS
Content-based image retrieval, Color Histogram, Relevance Feedback, k-means.
1. INTRODUCTION
Content-based image retrieval [1,18] has been an active research area in recent years. There has
been an experimental increase in the amount of non-text based data being generated from various
sources. In particular, images have been gaining popularity as an alternative and some time more
viable option for information storage [6]. Through the increase in storage and transmission
abilities more visual information is being made available on-line. These need to search and well
manage large volumes of Multimedia information [3]. However, while this presents a wealth of
information, it also causes a great problem in retrieving appropriate and relevant information
from databases.

10
Fig. 1 General model of CBIR
CBIR extracts low-level features which is Inbuilt in the images to present the contents of
images.Each image has features such as categories into three main classes: color [2, 11], texture
[9, 11] and shape [2, 10] features. This has resulted in a growing interest, and great active
research, into the extraction of relevant information from non text-based database. One of the
most common data in multimedia systems is image. Therefore, the capability of managing
images, allocating their quick and efficient indexing and retrieval are repeatedly followed in
multimedia database management systems. Researchers have been focusing on different ways of
retrieving information from database or objects based on their contents. Especially, content-based
image retrieval (CBIR) systems have attracted significant research and commercial interest. In
field of medicals, academic world, government and huge collections of images are being
produced. Many of these groups are the product of digitizing existing collections of analogue
photographs, diagrams, drawings, paintings, and prints.
Color is most common feature of an image such as used in Content-Based Image Retrieval [2, 9,
13]. These features are able to identify objects [10] and retrieve similar images on the basis of
their contents. These methods do work very efficient in object recognition and Web searching
[14].An efficient query reformulation is necessary for finding the relevant images from the
database. Relevance feedback (RF) is an interactive process which filters the retrieval results to a
particular query by utilizing the user’s feedback on previously retrieved images [5, 8].
II. Feature Extraction
Feature (content) extraction is the basis of content-based image retrieval. In sense, features may
include both visual features (color, texture, shape) and text-based features (key words,
annotations).
Color
Color feature is one of the most reliable and easier visual features used in image retrieval. It is
robust to background complication and is independent of image size and orientation [3].A lot of
techniques available for retrieving images on the basis of color similarity from image database
[1]. Each image included to the collection is analyzed to compute a color histogram which shows
the proportion of pixels of each color within the image. The color histogram for each image is
then stored in the database. Color, texture and shape information have been the primitive image
descriptors in content based image retrieval systems [7].
Feature
Extraction
Query
Image
Image
Collection
Feature
Database
Similarity
Measure
Match
Result
Feature
Query
Feature
Extraction
Relevance
Feedback

11
Texture
Textures are characterized by differences in brightness with high frequencies in the image
spectrum. They are useful in distinguishing between areas of images with similar color similar
color (such as sky and sea, or water, grass). A variety of methods has been used for measuring
texture similarity; the best- established depend on comparing values of what are well-known as
second-order statistics estimated from query and stored images. Essentially, these estimate the
relative brightness of picked pairs of pixels from each image. From these it is possible to
measures the image texture such as contrast, coarseness, directionality and regularity [16] or
periodicity, directionality and randomness [11].
Shape
Shape is an essential feature for perceptual object recognition and classification of images in
content-based image retrieval. Shape representation is significant concern both in object
recognition and classification. That means an image has to be segmented before extracting most
shape features. The shape representation typically is divided into two categories; boundary-based
and region-based. Fourier descriptors are a representative for the boundary-based method while
moment invariants are characteristic of the region-based method [20].
III. Related Work
Color is common and important feature in image retrieval system. Color comparison between two
images would however be time consuming and difficult problem to overcome this problem they
introduced a method of reducing the amount of information. One way of doing this is by
quantizing the color distribution into color histograms [12]. Using color histogram easier way for
color distribution or they used histogram divide in to different classes for matching.
K-means algorithm is one of the most widely used clustering algorithms in spatial clustering
analysis [17]. It is easy and efficient. But is also having limitations: It is sensitive to the
initialization. It doesn’t perform well in global searching and is easy to get into local optimization
and the improved K-means is based on the classical method to make the process of optimization
more dependent [4, 15].
Many of image retrieval applications are based on color feature and shape feature [3].Estimating
local texture based on pixels of the intensity image and a fuzzy index to point out the presence of
major colors [11]. It is based on the texture co-occurrence matrix.
K-means clustering for the classification of feature set obtained from the histogram refinement
method [4]. Histogram refinement provides a set of features for proposed for Content Based
Image Retrieval (CBIR) [2]. They used global histograms for image retrieval, because of their
effectiveness and insensibility to minor changes, are broadly used for content based image
retrieval.
IV. Proposed Work
Each image has three features Color, Shape and Texture. For fast and improve Image retrieval
performance we are using color feature extraction. Using color feature extraction firstly we
converted color image into grey level, this is containing values from 0 to 255.

12
Figure1. Block diagram of algorithm.
Algorithm:
1. Input as RGB Image.
2. RGB image convert into gray image or Intensity Image:
Intensity or gray image contains values from 0 to 255.
3. Gray Image converts into histogram bins sets
For i = 64 to 256
{
h(i) = g(xi,yi);
{
Distance = pdist(h(i));
Clust_distance = linkage [B, ‘centroid’];
%%measure Distance between clusters
Clust = cluster (c,’maxclust’,2);
%%Let (clust = = 1) be the coherent and (clust = = 2)
Incoherent cluster.
1
I = (R + G + B)
3
RGB
Image
Convert
into
Intensity
Image
Find
Clusters for
each
Bins-sets
Classify
Clusters as
coherent and
incoherent in
each Bins-set
Split into 4
Bins-sets
For each bins-sets
Calculate
1) Numbers of bins
coherent and incoherent
Basis
2) Avg Value of coherent
and incoherent pixels of
cluster
3) Percentage of coherent
and incoherent Pixels
cluster
For each coherent cluster
from every bins set calculate:
1) Size of Median in cluster
2) Size of smallest in cluster
3) Standard deviation of
Coherent pixels cluster

13
If (clust = = 1)
Call Calculate_num (n);
Call Calculate_Percent (Per,n);
Call Calculate_Avg (a,n);
Call Calculate_Median_bin(m,n);
Call Calculate_Smallest_bin(s,n);
Call Calculate_Std(t,n);
Else If (clust = = 2)
Call Calculate_num (n);
Call Calculate_Per(Per,n);
Call Calculate_Avg(a,n);
Else
“No cluster found”
End
6. Retrieval using query.
V. Result and Analysis
In this part, several tests are performing in order to analysis the performance of histograms and
here we used coral image database. To test the proposed method. Firstly RGB image converted to
intensity image. Then the image were split into bins sets and the features described in section IV
were calculated which is based on coherent and incoherent pixels. Followed these feature set;
images were grouped in similar clusters using k-means clustering method. Histogram refining
method more refines the histogram by dividing the pixels in a given container into a number of
classes based on color coherence vectors. Several features are calculated using Matlab[19] for
each of the cluster and these features are further classified using the k-means clustering method.
Table1. Example of parameter values for Incoherent pixels.

Binset (Xi) (CXi ) (µXi) (Yi) (CYi) (µYi)
1 52 1 51262 48 61 747
2 53 1 52065 47 127 364
3 51 1 50382 49 191 251
4 47 1 46207 53 255 204

14
We denoted Percentage of Coherent pixels (Xi), Number of Incoherent pixels clusters (CXi),
Average of Incoherent pixels cluster (µXi), Percentage of Incoherent pixels (Yi), Number of
Incoherent pixels clusters (CYi), Average of Incoherent pixels cluster (µYi).After calculation of
above features, we consider some additional features based on coherent bins cluster only. At this
stage, incoherent Pixels cluster is ignored. These features are preferred.
Similarly, we have additional parameter values related with the coherent bins in clusters including
size of median bins in cluster in each bin (MXi), size of smallest bins in cluster in each bin (SXi)
and standard deviation of coherent bins in clusters (DXi). for each of the largest, median and
smallest bin cluster value.
Table2. Additional features among the coherent Pixels cluster.
Two of them are based on the size of the Cluster. These intensity values median, standard
deviation, mean, variance, various sizes of objects are considered as appropriate features for
retrieval.
(a)Query Image
(b)Retrieval Result
Binset1 Binset2 Binset3 Binset4
(MXi) 324 161 91 76
(SXi) 0 0 2 0
(DXi) 4227700 913210 541820 386960

15
VI. Conclusion
This algorithm is based on color and texture features of static image. The intensity values median,
standard deviation and various sizes of the intensity values are considered as appropriate features
for retrieval. The proposed technique present very little amount of memory for features storage
and a prominent rate of computation and give good results in terms of accuracy. We have shown
that k-means clustering is fairly useful for appropriate image retrieval queries. The K-means
clustering algorithms to group the images content into different clusters based on the color feature
and k-means clustering algorithms have been often used in the pattern recognition. Animated
images performance behaviour is different based on this method.
ACKNOWLEDGEMENTS
First and foremost, I would like to express my sincere thanks to my guide Prof. Nishchol Mishra his wisdom and
abundance of encouragements from the very early stage of this research as well as giving me constant support
throughout the work and I also give special thanks to Prof. Sanjeev Sharma Head, SOIT, RGPV for his great support
and motivating me to complete my work. I would also like to give special thanks to my friends. If anyone is to be
acknowledged first, they are my family member’s for their inseparable support and prayers My Papa, Mommy and
Both Sister’s. I express my deep sense of gratitude to them for their support and continuous encouragement.
References
[1]. Bing Wang et al.”A Semantic Description For Content-Based Image Retrieval” College Of Mathematics And
Computer Science, Hebei University, Baoding 071002, China -2008.
[2]. Youngeum An et al. “Classification of Feature Set Using K-Means Clustering From Histogram Refinement
Method” (Korea Electronics Technology Institute, South Korea)- –2008.
[3]Waqas Rasheed et al. ”Sum of Values of Local Histograms for Image retrieval” - Chosun University, Gwangju,
South Korea-2008.
[4]. Tapas Kanungo et al. “An Efficient K-Means Clustering Algorithm: Analysis AND Implementation” Senior
Member, Ieee-2002.
[5]. Samar Zutshi et al. “Proto-Reduct Fusion BASED Relevance Feedback IN Cbir”, (Monash University)–-2009.
[6].James Z. Wang et al.``SIMPLIcity: Semantics-sensitive Integrated Matching for Picture LIbraries,'' IEEE Trans. on
Pattern Analysis and Machine Intelligence- 2001.
[7]. P. S. Hiremath et al.-“Content Based Image Retrieval based on Color, Texture and Shape features using Image and
its complement” -Dept. of P.G. Studies and Research in Computer Science, Gulbarga University,Gulbarga, Karnataka,
India-2007.
[8]. P. Suman Karthik et al. “Analysis Of Relevance Feedback In Content Based Image Retrieval” Center For Visual
Information Technology International Institute Of Information Technology Gachibowli, Hyderabad-India, -2006.
[9]. Zhi-Gang Fan et al “Local Patterns Constrained Image Histograms For Image Retrieval” Advanced R&D Center
Of Sharp Electronics (Shanghai) China -2008.
[10]. Tang li et al. “Developing a Shape-and-Composition CBIR Thesaurus for the Traditional Chinese Landscape”
University of Maryland College of Information Science College Park, MD, United States Library Student Journal, July
2007
[11]. Sanjoy Kumar Saha et al. “CBIR Using Perception Based Texture And Colour Measures”CSE Department; CST
Department Jadavpur Univ., India; B.E. College, Unit ISI, Kolkata, India - -2003.

16
[12]. Greg Pass Ramin Zabih et al. “Histogram Refinement for Content-Based Image Retrieval” Computer Science
Department Cornell University-1996.
[13]. Mohamad Obeid’ Bruno et al. ”Image Indexing & Retrieval Using Intermediate features”
Equipe MIRE ENIC, Telecom Lille , Cite Scientifique, Rue G. Marconi, 59658 Villeneuve d’Ascq France-2000.
[14]. M.S. Lew et al. “Content-Based Multimedia Information Retrieval: State Of The Art And Challenges” ACM
Transactions On Multimedia Computing- 2006.
[15]. Zhe Zhang et al. “Improved K-means Clustering Algorithm”-(College of Automation, Northwestern Polytechnic
University, Xi’an 710072) -2008.
[16]. Tamura et al. “Texture Feautres Corresponding to Visual Perception”- IEEE Trans on system, Man and cyber 8-
460-472-1978.
[17]. Prof. Dr. K. Sakthivel et al. “Performance Enhancement In Image Retrieval Using Modified K-Means Clustering
Algorithm”- Journal of Mathematics and Technology, Rangasamy College of Technology, Tiruchengode, Tamilnadu
ISSN: 2078-0257, February, 2010.
[18]. Morteza Analoui et al.” Content-based Image Retrieval Using Artificial Immune System (AIS) Clustering
Algorithms”-proceeding of the international Multi conference of Engineers and computer Scientists 2011 VolI, IMECS
– India 2011.
[19]. Www.mathworks.com, MATLAB and Simulink for Technical Computing , 2008.
Authors
1. Jayant Mishra, Asst. Prof. UIT, RGPV, Bhopal. He has obtained M. Tech. Degree
in Computer Application & Technology branch from School Of Information
technology, RGPV, Bhopal in 2009. He is also done MCA from Technocrats
Institute of Technolgy, Bhopal in 2007. He has worked as Asst. Prof. In Department
of Computer Science, Acropolis Institute Research & Technology- Bhopal, Since
Dec 2009 to March 2011. He’s working as Asst. Prof. in UIT, Rajiv Gandhi
Technical University, Bhopal (M.P.) Since March 2011. His one international
Journal has been published in Dec’2009. His research area of interests are Multi
Media Data mining & Digital Image Processing.
2. Anubhav Sharma. He has obtained BE from JIT Borawan (M.P.)in 2007. He’s
perusing M. Tech. (final sem) Degree in Computer Science branch from RITS-
RGPV, Bhopal in 2011. His research area of interests is Digital Image Processing.
3. Prof. Kapil Chaturvedi. He has obtained MCA from SATI (Degree), Vidisha
(M.P) in 2008. He’s perusing Ph.D. from RGTU Bhopal. He has worked as lecturer
In Department of Computer Applications, SATI (Degree), Vidisha (M.P.) Since Aug
2008 to March 2011 He’s Working as Assistant Professor in Department of
Computer Applications, UIT, Rajiv Gandhi Technical University, Bhopal (M.P.)
Since March 2011. My Area of Interest is Data Mining & Digital Image Processing.
.

An Unsupervised Cluster-based Image Retrieval Algorithm using Relevance Feedback

More Related Content

What's hot (20)

Similar to An Unsupervised Cluster-based Image Retrieval Algorithm using Relevance Feedback (20)

More from IJMIT JOURNAL (20)

Recently uploaded (20)

An Unsupervised Cluster-based Image Retrieval Algorithm using Relevance Feedback