Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Your Federal Quarterly Tax Payments are due April 15th Get Help Now >>

A Content Based Approach for Image Retrieval from Relevance Feedback

VIEWS: 40 PAGES: 5

									                                                                                                         ISSN No. 2278-3083
                                                  Volume 1, No.3, July – August 2012
                                International Journal of Science and Applied Information Technology
                                     Available Online at http://warse.org/pdfs/ijsait01132012.pdf




        A Content Based Approach for Image Retrieval from Relevance Feedback

                                 A.V. Senthil Kumar, Director, Department of Computer Applications,
                                          Hindusthan College of Arts & Science,Coimbatore.
                                                    avsenthilkumar@yahoo.com

                                 J.Sivakami, Research Scholar, Department of Computer Applications,
                                          Hindusthan College of Arts & Science,Coimbatore.
                                                     sivakamimscss@gmail.com

ABSTRACT
                                                                   database and extracting the feature into the color, space, text
Content-Based Image Retrieval (CBIR), is mainly based on           into the pattern mining. Finally retrieves the image on auto
finding images of interest from a large image database using       retrieval and compare the performance for proposed system.
the visual content of the images. Most of the approaches to        While the performance of a number of clustering algorithms
image retrieval were text-based, where individual images           in image retrieval has been analyzed in existing paper and
had to be annotated with format. Existing works are based          compare its performance with that of the automatic
on the performance of a number of clustering algorithms in         feedback.
image retrieval has been analyzed. The proposed work in
this paper is viewed on a new fuzzy based c-means                  Points in other clusters [4]. Clustering is a method of
partitional clustering algorithm. Partitional clustering           unsupervised classification, where data points are grouped
algorithm is used to improve the Content Based Image               into clusters based on their similarity. The goal of a
Retrieval and for comparing the performance of the image.          clustering algorithm is to maximize the intra-cluster
                                                                   similarity and minimize the inter-cluster similarity [3].
Keywords: Clustering, Partitional algorithm, CBIR, Fuzzy
C- means                                                           Partitional and hierarchical clustering are the most widely
                                                                   used forms of clustering. In partition clustering, the set of n
1.INTRODUCTION                                                     data points are partitioned into k non-empty clusters, where
                                                                   k ≤ n. In the case of hierarchical clustering, the data points
Knowledge discovery in databases process or KDD is                 are organized into a hierarchical structure [6], resulting in a
relatively young and interdisciplinary field of computer           binary tree or dendogram. In this paper, we propose a new
science is the process of discovering new patterns from            clustering algorithm, which would come under the category
large datasets involving methods at the intersection of            of partitional clustering algorithms. Two commonly used
artificial intelligence, machine learning, statistics and          methods for partitioning data points include the k-means
database system [1]. The goal of data mining is to extract         method and the k-medoids method. In the k-means method,
knowledge from a data set in a human-understandable                each cluster is represented by its centroid or the mean of all
structure. Data mining is the entire process of applying           data points in the cluster [5]. In the case of the k-medoids
computer-based methodology, including new techniques for           method, each cluster is represented by a data point close to
knowledge discovery [2], from data. Databases, Text                the centroid of the cluster. Apart from these methods, there
Documents, Computer Simulations, and Social Networks               has been lots of work on fuzzy partitioning methods and
are the sources of data for mining.                                partition methods for large scale datasets [8].

A cluster is a collection of data points that are similar to one   2. RELATED WORKS
another within the same cluster and dissimilar to data             Cluster analysis or clustering is the task of assigning a set of
content. The proposed approach for the image retrieval             objects into groups (called clusters) [7] so that the objects in
system which request a number of iterative feedbacks to            the same cluster are more similar (in some sense or another)
produce refined search results in a large scale image              to each other than to those in other clusters [4]. Clustering is
                                                                   a main task of explorative data mining, and a common
                                                                   technique for statistical data analysis used in many fields,

                                                                                                                                 65
@ 2012, IJSAIT All Rights Reserved
A.V.Senthil Kumar et al., International Journal of Science and Advanced Information Technology, 1 (3), July – August, 65-69




including machine learning, pattern recognition, image
analysis[8], information retrieval, and bioinformatics.                                  k-means algorithm. They define the contribution of a data
Clustering is a data mining (machine learning) technique                                 point belonging to a cluster as the impact that it has on the
used to place data elements into related groups without                                  quality of the cluster. This metric is then used to obtain an
advance knowledge of the group definitions [3]. Popular                                  optimal set of ‘k’ cluster from the given set of data points.
clustering techniques include k-means clustering and
Expectation Maximization (EM) clustering.                                                They now outline the proposed contribution-based
                                                                                         clustering algorithm. It optimizes on two measures, namely
A clustering is essentially a set of such clusters, usually                              the intra cluster dispersion given by
containing all objects in the data set [7]. Additionally, it
may specify the relationship of the clusters to each other,
for example a hierarchy of clusters embedded in each other.
Clustering can be roughly distinguished in:
                                                                                                     And the inter-cluster dispersion given by
           Hard clustering: each object belongs to a cluster or
            not
           Soft clustering (also fuzzy clustering): each object
            belongs to each cluster to a certain degree (e.g. a                          where k is the number of clusters and is the mean of all
            likelihood of belonging to the cluster)                                      centroids. The algorithm tries to minimize α and maximize
                                                                                         β.
Clustering algorithm. While the k-means algorithm
optimizes only on the intra-cluster similarity, our algorithm                            We use the notion of ‘contribution of a data point’ for
also The learning-enhanced feedback has been one of the                                  partitional clustering. The resultant algorithm requires only
most active research areas in content-based image retrieval                              three passes and we show that the time complexity of each
in recent years[8]. However, few methods using the                                       pass is same as that of a single iteration of the k-means
feedback are currently available to process relatively                                   clustering algorithm. While the k-means algorithm
complex queries on large image databases. In the case of                                 optimizes only on the intra-cluster similarity, our algorithm
complex image queries[9], the feature space and the                                      also optimizes on the inter-cluster similarity. Clustering has
distance function of the user’s perception are usually                                   widespread applications in image processing. Color-based
different from those of the system. This difference leads to                             clustering techniques have proved useful in image
the representation of a query with multiple clusters (i.e.,                              segmentation [9]. The k-means algorithm is quite popular
regions) in the feature space. Therefore, it is necessary to                             for this purpose. Clustering based on visual content of
handle disjunctive queries in the feature space. In this paper,                          images is an area that has been extensively is in research for
we propose a new content-based image retrieval method                                    several years. This finds application in image retrieval.
using adaptive classification and cluster merging to find
multiple clusters of a complex image query [10].                                         Content-Based Image Retrieval (CBIR) aims at finding
                                                                                         images of interest from a large image database using the
3. EXISTING SYSTEM                                                                       visual content of the images. Traditional approaches to
                                                                                         image retrieval were text-based, where individual images
Clustering is a form of unsupervised classification that aims                            had to be annotated with textual descriptions. Since this is a
at grouping data points based on similarity. In this paper,                              tedious manual task, image retrieval based on visual content
they propose a new partitional clustering algorithm based on                             is very essential [10].
the notion of ‘contribution of a data point’. They apply the
algorithm to content-based image retrieval and compare its                               At each feedback, the results are presented to the user and
performance with that of the k-means clustering                                          the related browsing information is stored in the log
algorithm.Partitional clustering aims at partitioning a group                            database. After accumulating long-term users’ browsing
of data points into disjoint clusters optimizing a specific                              behaviors, offline operation for knowledge discovery is
criterion. When the number of data points is large, a brute                              triggered to perform navigation pattern mining and pattern
force enumeration of all possible combinations would be                                  indexing. The log database maintains the separate details for
computationally expensive. Instead, heuristic methods are                                each log user and the feedback for each user and it recovers
applied to find the optimal partitioning. The most popular                               image at retrieval. These logs keep records of database
criterion function used for partitional clustering is the sum                            changes. If a database needs to be restored to a point beyond
of squared error function given by A widely used squared-                                the last full, offline backup, logs are required to roll the data
error based algorithm is the k means clustering algorithm. In                            forward to the point of failure.
this paper, we propose a clustering algorithm similar to the




                                                                                                                                                        66
@ 2012, IJSAIT All Rights Reserved
  A.V.Senthil Kumar et al, International Journal of Science and Advanced Information Technology, 1 (3), July – August, 65-69



Computing distance measures based on color similarity is
achieved by computing a color histogram for each image
that identifies the proportion of pixels within an image
holding specific values (that humans express as colors).
Current research is attempting to segment color proportion
by region and by spatial relationship among several color
regions. Examining images based on the colors they contain
is one of the most widely used techniques because it does
not depend on image size or orientation. Color searches will
usually involve comparing color histograms, though this is
not the only technique in practice.

4. PROPOSED SYSTEM

The proposed approach for the image retrieval system
which request a number of iterative feedbacks to produce
refined search results in a large scale image database and
extracting the feature into the color, space, text into the
pattern mining. High quality of image retrieval on RF can
be achieved in a small number of feedbacks. To resolve the
problems existing in current RF, such as redundant
browsing and exploration convergence. The approximated
solution takes advantage of exploited knowledge
(navigation patterns) to assist the proposed search strategy
in efficiently hunting the desired images. All positive
images are considered for navigation pattern mining. It
focuses on the discovery of relations among the users’
browsing behaviors on RF. the frequent patterns mined
from the user logs are regarded as the useful browsing paths
to optimize the search direction on RF. The navigation
pattern mining patterns from the user log in the log database
and indexing the pattern .It finally retrieves the image on                             5. EXPERIMENTAL RESULTS
auto retrieval and compare the performance for proposed
system. The concept of contribution to find the optimal                                 The images were clustered using our algorithm with the
cluster number, we use it in a different manner for optimal                             initial centroids chosen at random. The cluster whose
partitioning of the data points into a fixed number of                                  centroid was closest in distance to the given test image was
clusters.                                                                               determined and the images belonging to the cluster were
                                                                                        retrieved. The results were then compared with images
                                                                                        retrieved using the kmeans clustering algorithm with the
                                                                                        same set of initial centroids.

                                                                                        In this Figure 1, the proposed images basic partitional
                                                                                        clustering algorithm for Content-Based Image Retrieval is
                                                                                        compared with the proposed Fuzzy algorithm partitional
                                                                                        clustering for Content-based image retrieval. When the
                                                                                        number of clusters increases, the average precision will be
                                                                                        increased. The average precision rate varies in the interval
                                                                                        of 0.1. Fuzzy partitional clustering algorithm gives better
                                                                                        results than the existing clustering algorithm. The number
                                                                                        of clusters increased by 1.




                                                                                                                                                  67
@ 2012, IJSAIT All Rights Reserved
A.V.Senthil Kumar et al, International Journal of Science and Advanced Information Technology, 1 (3), July – August, 65-69



                                                                                             6. CONCLUSION

                                                                                            In this paper, the basic partitional clustering algorithm for
                                                                                            Content-based image retrieval is compared with the
                                                                                            proposed fuzzy algorithm partitional clustering for Content-
                                                                                            based image retrieval. When the number of clusters
                                                                                            increases, the Cluster accuracy will increased. The main
                                                                                            feature RF is to efficiently optimize the retrieval quality of
                                                                                            interactive CBIR. On one hand, the patterns derived from
                                                                                            the users’ long term browsing behaviors are used as a good
                                                                                            support for minimizing the number of user feedbacks. On
                                                                                            the other hand, the proposed algorithm RF Search performs
                                                                                            the pattern-based search to match the user’s intention by
                                                                                            merging three query refinement strategies. The proposed
                                                                                            approach for the image retrieval system which request a
                                                                                            number of iterative feedbacks to produce refined search
   Figure 1: Comparison of K-means, FCM and Proposed                                        results in a large scale image database and extracting the
   method                                                                                   feature into the color, space, text into the pattern mining.
                                                                                            High quality of image retrieval on RF can be achieved in a
   In this Figure 2, we are taking the two parameters as                                    small number of feedbacks. To resolve the            problems
   methods and retrieval accuracy. In that the X axis represents                            existing in current RF, such as redundant browsing and
   the methods parameter and Y axis denotes the retrieval                                   exploration convergence. The approximated solution takes
   accuracy.                                                                                advantage of exploited knowledge (navigation patterns) to
                                                                                            assist the proposed search strategy in efficiently hunting the
                                                                                            desired images. In the future, integrate user’s profile into
                                                                                            NPRF to further increase the retrieval quality. We will
                                                                                            apply the NPRF approach to more kinds of applications on
                                                                                            multimedia retrieval or multimedia recommendation Present
                                                                                            a set of experiments to evaluate the performance of the
                                                                                            proposed approach with the existing. The result also shows
                                                                                            that this enhanced approach performs better than
                                                                                            conventional techniques.

                                                                                            7. REFERENCES

                                                                                            1. M.D. Flickner, H. Sawhney, W. Niblack, J. Ashley, Q.
                                                                                            Huang, B.Dom, M. Gorkani, J. Hafner, D. Lee, D. Steele,
                                                                                            and P. Yanker. Query by Image and Video Content: The
        Figure 2: Average precision and Number of Clusters                                  QBIC System, Computer, Vol. 28, No. 9, pp. 23-32, Sept.
                                                                                            1995.
    S.No                       Method                             Retrieval
                                                                 Accuracy %                 2. R. Fagin. Combining Fuzzy Information from
       1        K-means                                            80-85%                   Multiple Systems, Proc. Symp. Principles of Database
       2        Fuzzy C- means                                     85-90%                   Systems (PODS), pp. 216-226, June1996.
       3        Proposed system                                    90-95%
                                                                                            3. R. Fagin. Fuzzy Queries in Multimedia Database
                                                                                            Systems, Proc.Symp. Principles of Database Systems
   In our paper image retrieval is used as the parameter to find                            (PODS), pp. 1-10, June 1998.
   how much level of accuracy is achieved in image retrieval
   .The retrieval of images is compared with the three                                      4. J. French and X-Y. Jin, An Empirical Investigation of
   algorithm such as K means, Fuzzy c means (FCM) and                                       theScalability of a Multiple Viewpoint CBIR System,
   proposed method From the graph shows that, the Fuzzy c                                   Proc. Int’l Conf.Image and Video Retrieval (CIVR), pp.
   means (FCM) has better than the retrieval accuracy of the                                252-260, July 2004.
   image rather than the K means but the proposed system has
   high level of the image retrieval accuracy than other two
   methods. Finally the retrieval of the image accuracy is high
   in the proposed system.




                                                                                                                                                        68
   @ 2012, IJSAIT All Rights Reserved
A.V.Senthil Kumar et al, International Journal of Science and Advanced Information Technology, 1 (3), July – August, 65-69




5 D. Harman. Relevance Feedback Revisited, Proc. 15th                                  8. D.H. Kim and C.W. Chung. Qcluster: Relevance
Ann. Int’lACM SIGIR Conf. Research and Development in                                  Feedback UsingAdaptive Clustering for Content-Based
Information Retrieval, pp. 1-10, 1992.                                                 Image Retrieval, Proc.ACM SIGMOD, pp. 599-610, 2003.

6. Y. Ishikawa, R. Subramanya, and C. Faloutsos.                                       9. K. Porkaew, K. Chakrabarti, and S. Mehrotra, Query
MindReader: Querying Databases through Multiple                                        Refinementfor Multimedia Similarity Retrieval in
Examples, Proc. 24th Int’lConf. Very Large Data Bases                                  MARS, Proc. ACM Int’lMultimedia Conf. (ACMMM), pp.
(VLDB), pp. 218-227, 1998.                                                             235-238, 1999.

7. Jin and J.C. French. Improving Image Retrieval                                      10.J. Liu, Z. Li, M. Li, H. Lu, and S. Ma. Human
Effectivenessvia Multiple Queries, Multimedia Tools and                                BehaviourConsistent Relevance Feedback Model for
Applications, vol. 26,pp. 221-245, June 2005.                                          Image Retrieval, Proc.15th Int’l Conf. Multimedia, pp.
                                                                                       269-272, Sept. 2007.




                                                                                                                                           69
@ 2012, IJSAIT All Rights Reserved

								
To top