International Journal of Computer Science Engineering

and Information Technology Research (IJCSEITR)

ISSN(P): 2249-6831; ISSN(E): 2249-7943
Vol. 5, Issue 1, Feb 2015, 35-42
TJPRC Pvt. Ltd.

INTERACTIVE IMAGE RETRIEVAL WITH ADVANCED CLUSTERING STRATEGY

ANUJA KHODASAKR & SIDDHARTH LADHAKE
SIPNA COET, Amravati University, Maharashtra, India

ABSTRACT
This paper presents an image retrieval engine based on advanced clustering strategy which improves performance
of image retrieval in term of precision, recall, accuracy and retrieval time. This paper also highlights an overview of
recently developed image clustering techniques and their real time applications such as Clustering based image binarization
in palm leaf manuscripts, Blood Vessel Extraction for retinal images, synthetic aperture radar images, tumor detection in
MRI images etc. Clustering is usually used in image processing, data mining, image retrieval, pattern recognition, image
segmentation etc. Clustering provides an effective analysis and proper validation of image for retrieval. The proposed
system is employed advanced clustering strategy which utilized N-cut clustering, hierarchical and fuzzy k-means clustering
techniques. Cluster based image retrieval system search similar images based on clusters that calculated by color, texture
and shape and compare similarity with query image clustered feature. Clustered based image retrieval is fast and accurate.
Here, K mean clustering form clusters of similar image features and then N-cut arrange these clusters to improve retrieval
speed. Hierarchical clustering is applied on retrieved set of images in relevance feedback to improve accuracy. Proposed
content based image retrieval system based on novel clustering strategy significantly improves performance of image
retrieval. Experimental result reveals that there is outstanding improvement in accuracy, retrieval time, precision and recall
of the image retrieval system.

KEYWORDS: Clustering Strategy, Hierarchical Clustering, Image Retrieval Engine, K-Mean Clustering, N-Cut
Clustering, Precision

INTRODUCTION
In general, cluster analysis is an analysis for grouping a set of objects in such a way that objects in the same group.
Such group is called a cluster. It is soul of data mining and statistical data analysis. it is integral part of many fields such as
machine learning, pattern recognition, image analysis, information retrieval, and bioinformatics. Cluster analysis is a
general task which carried out by one or more algorithms in combination. Some popular clustering algorithms are
connectivity based clustering or hierarchical clustering, centroid-based clustering, distribution-based clustering, and
density-based clustering. all clustering techniques are implemented in many real life applications such as biology,
computational biology and bioinformatics, plant and animal ecology, sequence analysis, high- throughput genotyping
platforms, human genetic clustering, medicine, medical imaging, IMRT segmentation, social network analysis, search
result grouping etc.

ADVANCED DEVELOPMENT IN CLUSTERING TECHNIQUES

This segment gives a overview of current developed image clustering techniques such as an Adaptive Memetic
Fuzzy Clustering Algorithm, Local learning based image clustering model, hierarchical clustering with the correlation

www.tjprc.org

editor@tjprc.org

36

Anuja Khodasakr & Siddharth Ladhake

distance measure, Clustering by Composition etc.

Remote sensing imagery is clustered with an adaptive memetic fuzzy clustering algorithm with spatial
information with high clustering quality [1]. This algorithm defines and optimize a objective function which combining
differential evolution algorithm with a local search method using Gaussian local search. Local learning
based image clustering model which is more effective and efficient, used for discriminant analysis where local discriminant
model evaluate image clustering at local level, and the optimal image features were obtained using image interpolation
approach [2]. In this clustering model, image feature reduction is performed through two-dimensional PCA by extracting
significant eigenvectors of the image dataset. Clustering by Composition is state-of-the-art image clustering techniques [3],
which composes multiple images from each other simultaneously and efficiently by using a collaborative randomized
search algorithm. This algorithm helps in unsupervised detection of very challenging image categories.
Consistent Color of Contour is preserved by image clustering technology based on involved spatial distance which
assigns the contour a consistent color [5]. Here, the contour will be cluster in many clusters bring by the color gradient of
the contour. The hierarchical clustering with the correlation distance measure has the best performance in low quality Farsi
sub-words for word recognition to reduce the number of classes of sub-words in order to improve the overall recognition
rate and this clustering method has high performance than other clustering methods and distance measures [10]. This new
method for clustering exploited two clustering methods, hierarchical and k-means and use Local binary patterns and zoning
algorithms for feature extraction. Prediction Error Clustering and Random Permutation are used to design an
Efficient Image Encryption-Then-Compression System by considering both lossless and lossy compression [11]. A novel
unsupervised image-based SPM model based on the fuzzy c-means clustering approach was presented where Subpixel
mapping obtain a land cover map with finer spatial resolution. This system generates higher accuracy subpixel land cover
maps [13]. Sparse Riemannian Manifold Clustering is deal with the problem of segmenting high angular resolution
diffusion imaging data into multiple regions with distinct diffusion properties [14]. The orientation distribution function is
use to model diffusion and cast the ODF segmentation problem as a clustering problem in the space of ODFs.

SIGNIFICANCE OF CLUSTERING IN IRS

Objective of clustering is uniformly partitioning feature space of feature vectors in a significant manner for the
application domain context. A cluster is a collection of similar objects and dissimilar objects belonging to other clusters.
Subsequently, the aim of clustering is to establish the intrinsic grouping in a set of unlabeled data. Clustering perform
nearest neighbor search effectively. Image retrieval system is fatally suffered by semantic gap problem which address by
using unsupervised learning based on clustering strategies. Cluster-based image retrieval approach retrieval relevant cluster
of image to query image with reasoning for selection of cluster. This retrieval system is based on similarity driven image
retrieval which design virtually any symmetric real-valued image similarity measure. Clustering technique present a
dynamic and local visualization of the image database where clusters are created depending on a query image.
Figure 1 shows proposed clustering strategy.

Impact Factor (JCC): 6.8785

Index Cope rnicus Value (ICV): 3.0

Interactive Image Retrieval with Advanced Clustering Strategy

37

.
Figure 1: Proposed Clustering Strategy

ADVANCED CLUSTERING APPLICATIONS

Various latest application of clustering algorithms and techniques are highlighted in this section such as automatic
recognition of follicles in ultrasound images, synthetic aperture radar image change detection job, detection of tumor in
MRI of brain images etc.
A novel method for automatic recognition of follicles in ultrasound images which diagnosis infertility in women,
are presented using Discrete wavelet transform based K-means clustering [12]. A data-driven probabilistic structure
performs atlas-guided segmentation of a heterogeneous set of brain MR images and clusters the images in homogeneous
subgroups, while constructing separate probabilistic atlases for each cluster to direct the segmentation [4]. An effective
unsupervised approach based on the combined difference image and k-means clustering is proposed for the synthetic
aperture radar image change detection job [7]. In the proposed system, the k-means clustering algorithm is used to cluster it
into two classes, changed area and unchanged area. Genetic c-means and k-means clustering techniques which provide fast
and efficient clustering results are used to detect tumor in MRI of brain images [8]. An automated blood vessel detection
method from the fundus image is successfully implemented by using morphological operator and KCN clustering [15].
Clustering techniques is successfully used for many agriculture applications. The concept of clustering is
applied in palm leaf image binarization with three dimensional features to accurately extract foreground information from
palm leaf images [16]. This system uses a set of ground truth corresponding to ten palm leaf images and retrieves and
preserves the historical knowledge.

FRAMEWORK OF PROPOSED SYSTEM

Proposed cluster based image retrieval system search images on the basis of clusters that calculated on color,
texture and shape as shown in figure 2. Proposed clustering strategy utilized K-mean clustering, hierarchical clustering and
N-cut clustering to improve speed and accuracy of image retrieval. At first, image are group in clusters by using K mean
and then N-cut arrange these clusters to improve retrieval speed. Hierarchical clustering is used in relevance feedback
retrieval to improve accuracy.
Image content such as colour, texture and shape features are extracted and then segments to get objects. Texture
features are obtained by Haar wavelet transform which is discontinuous step function. Colour feature are extracted using
histogram. For segmenting an image, fuzzy K-Means clustering algorithm group similar pixel together into k groups with
cluster centers and select the value of k of K-Means dynamically to determined the number of k values instead of randomly

www.tjprc.org

editor@tjprc.org

38

Anuja Khodasakr & Siddharth Ladhake

chosen. Fuzzy k-mean clustering extended to fuzzy partitioning of data points. As mention earlier, the proposed contentbased image retrieval system utilised color, texture and shape content of image. The module is related to K-Means
processing and final image grouping for retrieval. The resultant extracted features are placed in k clusters by K-Means
methodology and then all k clusters are arranged by n-cut method. Average linkage method computes the distance between
the new cluster and remaining clusters. Objects are extracted from segments and clustered by using N-cut. Similar objects
are grouped in same cluster. Object clustering increase the speed of image retrieval system. Each object has six features,
which are the average features of all the member pixels. K-mean processing takes k parameters and partitions a set of n
objects into k clusters hence resulting intra-cluster similarity is high. An object is assigned to the most similar cluster.
This object assignment is based on the distance between the object and the cluster center closest to it from K cluster.
It calculates the new-fangled centroid and in this way each center finds the centroid of its own points. N-cut method
presents globally optimized clustering which allows a low-dimensional image clustering with significant reduction of the
complexity.
When user fired query in the form of image, then, the query image goes through the same image segmentation
algorithm to obtain objects. These objects are compared to the cluster centers in the database and similarity is calculated
using L2 distance. The objects in the database that has a minimum distance will be returned to perform global image
distance computation between query image and database image. The objects in the database that has a minimum distance
will be returned to perform global image distance computation between query image and database image.
Relevance feedback based on hierarchical clustering strategy is employed to improve system performance by
filtering retrieved images. Then, images are hierarchically grouped by their similarity levels. Hierarchical clustering return
outputs in the form of hierarchy, a structure that is more informative than the unstructured set of clusters. This results in
random groups of images which are further organised by n-cut. Proposed method uses a hierarchical cluster representation
of the relevant and irrelevant images for given query image. This strategy applies on first initial set of the retrieved images.
In this process, primarily the image groups that are most similar are clustered together. Then the images in the groups are
sorted in a descending order based on their similarity values and find out most similar to the query image.

Figure 2: Proposed Content Based Image Retrieval System Based on Clustering

Impact Factor (JCC): 6.8785

Index Cope rnicus Value (ICV): 3.0

Interactive Image Retrieval with Advanced Clustering Strategy

39

EXPERIMENTAL RESULT
Empirical result shows image retrieval performance improves in term of retrieval time, precision and accuracy.
We performed testing of proposed image retrieval system by using standard databases. The retrieval system take query
image as input and gives set of images as output. For initial testing, we use four images, white rose flower, using color,
texture and shape feature for retrieval. It also employed fuzzy k-means clustering and N-cut algorithm for images retrieval.
For analysis performance of image retrieval system with proposed clustering strategy, we set up a series of test on the
clustering performance.
In Figure 3, the query image is a white rose flower and feature vector composed of color, shape and texture.
Proposed clustering strategy for the images retrieval can remove some images that are visually irrelevant to the query
image for improving accuracy of retrieval result.

Figure 3: Proposed System Input as Query Image and Output as Set of Retrieved Image
Eperimental result show performance evaluation in term of precision, recall and accuracy which given by
equation (1), (2), and (3) respectively.

(1)

(2)

(3)

www.tjprc.org

editor@tjprc.org

40

Anuja Khodasakr & Siddharth Ladhake

Table 1: Result Evaluations

Query
Images

Numeric Evaluation of Retrieval Result

Accuracy
Precision
Recall
88.66

89.45

89.05

79.23

85.63

82.30

86.56

89.60

88.05

89.55

90.55

90

Table 1 shows performance evaluation in term of three parameter such as precision, recall and accuracy. Fourth
parameter, retrieval time is reduced by using proposed clustering strategy which demonstrated by experimental assessment
on the content based image retrieval system with set of four test query images. At first, we perform image retrieval without
proposed clustering strategy and with clustering strategy. Empirical result clearly shows that retrieval time is reduced
significantly with proposed hybrid clustering approach as shown in figure 4. Figure 5 and figure 6 shows screen shot for
clustering asigment and clustering result for proposed clustering strategy.

Figure 4: Time Required for Image Retrieval with and without Proposed Clustering Strategy

Figure 5: Clustering Assigment

Impact Factor (JCC): 6.8785

Index Cope rnicus Value (ICV): 3.0

Interactive Image Retrieval with Advanced Clustering Strategy

41

Figure 6: Proposed Clustering Result for White Rose Flower Query

CONCLUSIONS

Content based image retrieval system based on novel clustering strategy composed of k-mean clustering,
hierarchical clustering and N-cut clustering which improve performance of image retrieval.

Experimental evaluation carried on standard image data set. Experimental result reveals that there is outstanding
improvement speed, accuracy, precision and recall of the image retrieval system.

Proposed advanced clustering techniques plays crucial role in improvement of searching and retrieval of images in
many real world applications such as medical applications and space research based applications.

REFERENCES
1.

Yanfei Zhong, Ailong Ma, Liangpei Zhang. (2014). An Adaptive Memetic Fuzzy Clustering Algorithm with
Spatial Information for Remote Sensing Imagery. Selected Topics in Applied Earth Observations and Remote
Sensing, (pp. 1235 1248). IEEE Journal.

2.

Ahmed N, Jalil A. (2014). Image clustering using local discriminant model and two-dimensional PCA features.
Applied Sciences and Technology, pp.145 149, IEEE International Conference.

3.

Faktor A, Irani M. (2014). Clustering by Composition.Unsupervised Discovery of Image Categories. Pattern

Analysis and Machine Intelligence, (pp. 1092 1106). IEEE Transaction.

4.

Ribbens A, Hermans J, Maes F, Vandermeulen D, Suetens P. (2014). Unsupervised segmentation, Clustering, and
Group wise Registration of Heterogeneous Populations of Brain MR Images. Medical Imaging, (pp. 201 224).
IEEE Transaction.

5.

Yang, Sheng Chih, Huang, Shih Yen, Liu, Xi Ru. (2014). An Image Clustering Technology for Preserving the
Consistent Color of Contour. Computer, Consumer and Control, (pp.107 110). IEEE International Symposium.

6.

De-An Huang, Li-Wei Kang, Wang, Y.-C. F, Chia-Wen Lin.(2014). Self-Learning Based Image
Decomposition with Applications to Single Image Denoising. Multimedia, (pp. 83 93). IEEE Transaction.

7.

Yaoguo Zheng, Xiangrong Zhang, Biao Hou, Ganchao Liu. (2014). Using Combined Difference Image and

www.tjprc.org

editor@tjprc.org

42

Anuja Khodasakr & Siddharth Ladhake

-Means Clustering for SAR Image Change Detection. Geoscience and Remote Sensing Letters, (pp. 691 695)
IEEE.
8.

Sinha K, Sinha G. R.(2014). Efficient segmentation methods for tumor detection in MRI images. Electrical,
Electronics and Computer Science, (pp. 1 60) IEEE International Conference.

9.

Deliang Xiang, Tao Tang, Canbin Hu, Yu Li, Yi Su. (2014)A Kernel Clustering Algorithm With Fuzzy Factor:
Application to SAR Image Segmentation. Geoscience and Remote Sensing Letters, (pp. 1290 1294).
IEEE letter.

10. Yarmohammadi H. A, Fard A. A, Khosravi H. (2014). Clustering low quality Farsi sub-words for word
recognition. Intelligent Systems, (pp. 1 5). IEEE International Conference.
11. Jiantao Zhou, Xianming Liu, Au O. C, Yuan Yan Tang.(2014). Designing an Efficient Image Encryption-ThenCompression System via Prediction Error Clustering and Random Permutation. Information Forensics and
Security, (pp. 39 50). IEEE Transaction
12. Kiruthika V, Ramya M. M. (2014). Automatic Segmentation of Ovarian Follicle Using K-MeansClustering.
Signal and Image Processing, (pp. 137 141). IEEE International Conference.
13. Yihang Zhang, Yun Du, Xiaodong Li, Shiming Fang, Feng Ling. (2014). Unsupervised Subpixel Mapping of
Remotely Sensed Imagery Based on Fuzzy C-Means Clustering Approach,. IEEE, Geoscience and Remote
Sensing Letters, Vol. 11, No.5, pp. 1024 1028, 2014.
14. Cetingul H. E, Wright M. J, Thompson P. M, Vidal R. (2014). Segmentation of High Angular Resolution
Diffusion MRI Using Sparse Riemannian Manifold Clustering, Medical Imaging, (pp.301 317).
IEEE Transaction.
15. Mehrotra A, Tripathi S, Singh K. K, Khandelwal P. (2014). Blood Vessel Extraction for retinal images using
morphological operator and KCN clustering. Advance Computing Conference, (pp. 1142 1146).
IEEE International conference.
16. Krishna M. P, Sriram A, Puhan N. B. (2014). Clustering based image binarization in palm leaf manuscripts.
Advance Computing Conference, (pp.1060 1065). IEEE International conference.
17. Hertz, T, Shental, N, Bar-Hillel, A, Weinshall, D. (2003). Enhancing image and video retrieval: learning via
equivalence constraints. Computer Vision and Pattern Recognition, (pp. II - 668-740. IEEE International
conference.
18. Wenbing Tao, Wuhan, Hai Jin, Yimin Zhang. (2007) Color Image Segmentation Based on Mean Shift and
Normalized Cuts. Systems, Man, and Cybernetics, (pp.1382-1389). IEEE Transaction.

Impact Factor (JCC): 6.8785

Index Cope rnicus Value (ICV): 3.0