ISSN ONLINE(2319-8753)PRINT(2347-6710)

All submissions of the EM system will be redirected to Online Manuscript Submission System. Authors are requested to submit articles directly to Online Manuscript Submission System of respective journal.

Augmented Content Based Image Retrieval using Fusion Based on Self Organizing Map.

R.Priya1, Dr.VasanthaKalyani David2
Dept. of Computer Science, SreeNarayana Guru College, Avinashilingam University for Women, Coimbatore, India1,2
Corresponding Author: SHARMA VIVEK, E-mail: vivek03sharma@rediffmail.com
Related article at Pubmed, Scholar Google

Visit for more related articles at International Journal of Innovative Research in Science, Engineering and Technology

Abstract

Content Based Image Retrieval (CBIR) comprises the task of retrieving the image from a large database which matches the query image. The features like color, texture and shape are taken as multiple features and using fusion Self Organizing Map the matching process is carried out. Enhanced HSV-based Histograms for color, Active Contours for shape, DWT transformation for texture are applied and this work prove that the proposed CBIR system is an improved version in terms of precision, recall and speed of image retrieval.

Keywords

Content based image retrieval, enhanced HSV-based histograms, Active Contours, Discrete Wavelet Transform, Self Organizing Maps

INTRODUCTION

Day by day image database is expanding as digital images capturing has become very easy and is the cheapest. As the database grows retrieval from it becomes tedious. So image retrieval using text as its file name is not sufficient enough. So the concept of content based image retrieval (CBIR) emerged. CBIR concentrates on content like color, shape, texture and spatial features.
In spite of its wide popularity, the field has not yet reached its maturity and is not yet being used on a significant scale. The usage statistics shows a diversified opinion on the usefulness and effectiveness of CBIR systems in handling queries on large and diverse real-life image databases. This shows that the research area is still active and methods to improve existing solutions are in great demand.
A CBIR system is defined as a task which uses an amalgamation of techniques, like statistics, image processing and pattern recognition, with the aim of searching and retrieving images from a large image database that match a given query in an efficient manner. The query can either be an example image or a sketch, text or terms of keywords.
Traditionally, an image is represented by its features such as the color, texture and shape of objects. A great amount of work proposed use algorithms that perform similarity matching on any one of the various available features of the image. Only recently, the usage of multiple features for image retrieval is being gaining attention as in [9] and [15]. It has been discovered that the retrieval performance, in terms of recall and precision, is enhanced when multiple feature vectors are used. The reason for this performance increase is that single features extracted from images characterizes different aspects of an image content, which when combined to form multiple features, provide an adequate and more accurate description of image content as in [6]. CBIR system using multiple features extracts more than feature from the image and performs retrieval process in two manners. The first type merge the multiple feature vectors to create a global index while the second method create separate index structure for each feature and then merge the results. The result of fusion is then used during retrieval and indexing. This paper considers the first type, that is, feature level fusion

II. II.PROPOSED METHODS

A.Color Feature
Color feature is the most attractive one. Color histogram retrieval method is enhanced with Self Organizing Map (SOM) for dimensionality reduction and to improve accuracy. SOM uses improved histogram as image feature vector. Feature vector F(I) is generated for each image in the collection. When the query is made, the feature vector of query image is calculated and similarities between the two images are calculated using Euclidean distance. The weight vectors are stored using single color histogram. As each pixel stores three values of HSV color space to handle weight vectors of K dimensions the pixels are grouped in K/3 tiles. The weight vector factors are calculated using the equation
image
where j indicates the color levels of color space C, Wj is the jth weight of HR.
The image blocks are given as input to the network. These input vectors(X) are mapped with network weight vector (Wi). Then the similarity between X and Wi is measured in terms of spatial neighborhood Nm and is found best matching node m such that
||x – wm || = min [||x – wi||]. (1a)
The radius of Nm will be decreasing as the training progresses. Towards the end the neighborhood will have no cells other than the central winning one. The weight updating rule for self organizing feature map is defined as
Δwi(t) = α[x(t) – wi(t)] for i ? Nm(t) (1b)
whereNm (t) denotes the current spatial neighborhood and α denotes the learning rate. After training the weight vectors of each neuron of the Kohonen layer acts as code vectors.
B.Texture Features
To extract texture features standard DWT combined with Rotated wavelet filter [RWF] and dual tree complex wavelet transform [DT-WT] is used. For each sub band, 4 texture features are calculated they are energy, entropy, contrast & homogeneity.
image
image
image
image
Where Co(i, j) is the co-occurrence matrix and is computed for several values of displacements and the one which maximizes the statistical measure is used. The co-variance matrix and displacements are calculated for each color sub band and are used. The average value is used as the final texture feature set for an image.
C.Shape Features
For extracting shape features, active contour model as in[16,] is used to detect the various regions of an image and the shape features are retrieved using the method proposed in[17]. The energy function of the active contour model is defined by the following equation.
E*snake= ∫(Einternal[v(s)]+Eimage[v(s)] + Econstrant[v(s)]) ds (6)
The minimum of the energy function locates the final template of the active contour model.
D.Feature Fusion
Let C be the set of color features, T be the set of texture features and S be the set of shape features. Let R be the resultant feature vector after concatenating C, T and S. The average of all the respective features over the entire database is used to normalize the individual feature components. Let the normalized vectors be C', T' and S'. To avoid the curse of dimensionality, a k-means clustering algorithm is used. The k-means algorithm performs a two step procedure, where the first step removes redundant features and the second step retains points and removes points that are very near to the specific point. These points are removed because they may not provide additional information because of being in vicinity. The distance classifier used is Euclidean distance and the number of clusters is determined using the PBM cluster validity index as in [18]. To further select optimum feature vector, the SOM method proposed in [20] is used.
A self-organizing map (SOM) or self-organizing feature map (SOFM) is a type of artificial neural network that is trained using unsupervised learning to produce a low-dimensional (typically two-dimensional), discretized representation of the input space of the training samples, called a map. Self-organizing maps are different from other artificial neural networks in the sense that they use a neighborhood function to preserve the topological properties of the input space.
Like most artificial neural networks, SOMs operate in two modes: training and mapping. Training builds the map using input examples. It is a competitive process, also called vector quantization. Mapping automatically classifies a new input vector.
A self-organizing map consists of components called nodes or neurons. Associated with each node is a weight vector of the same dimension as the input data vectors and a position in the map space. The usual arrangement of nodes is a regular spacing in a hexagonal or rectangular grid. The self-organizing map describes a mapping from a higher dimensional input space to a lower dimensional map space. The procedure for placing a vector from data space onto the map is to first find the node with the closest weight vector to the vector taken from data space. Once the closest node is located it is assigned the values from the vector taken from the data space.
This paper is an extension of the authors’ previous work in [20] , where an enhanced CBIRsystem based on Self Organizing Map (SOM) and histogram was proposed. This system integrated the spatial information with enhanced color histograms. In this paper, this work is extended to consider multiple features, enhanced color histogram, texture and shape that are combined with SOM for dimensionality reduction. The initial work is referred to as ‘SOM-Color CBIR System’, while the proposed system is referred to as ‘Fusion based on SOM CBIR System’ in this paper.

III.EXPERIMENTAL RESULTS

A collection of around 650 images are taken randomly from the web as the image database. The system is developed in MATLAB7.3. The histograms for all the images were in 72 color bins in HSV color spam. The performance metrics taken for evaluation is precision-recall measure and retrieval time.
The result shows that Fusion SOM CBIR system is improved than the SOM Color CBIR. Thus the proposed system shows 14.37% of speed efficiency than SOM color CBIR.

IV. CONCLUSIONS

This paper considers the multiple features color, texture and shape along with Fusion based on Self Organizing Map CBIR for image retrieval. For color feature it uses enhanced histogram, for texture dual tree complex wavelet transform and for shape active contour model. The experimental result shows that this combination is more efficient than the previous one in terms of precision-recall. Further the research can be continued with multiple query images for retrieval. Further the efficiency can be improved by best method of feature selection.

Figures at a glance

Figure 1 Figure 2 Figure 3
Figure 1 Figure 2 Figure 3
 

References

  1. Xiaoling, W. (2009) A Novel Circular Ring Histogram for Content-Based Image Retrieval, First International Workshop on Education Technology and Computer Science, Vol. 2, Pp.785-788.
  2. Wang, S. and Qin, H. (2009) A Study of Order-Based Block Color Feature Image Retrieval Compared with Cumulative Color Histogram Method, Sixth International Conference on Fuzzy Systems and Knowledge Discovery, FSKD '09, IEEE Xplore, Vol. 1, Pp. 81 – 84.
  3. Pass, G. and Zabih, R, (1996) Histogram refinement for content-based image retrieval, Proc. WACV [C]. Sarasoto, FL, Pp. 96-102.
  4. Banerjee, M., Kundu, M.K. and Das, P.K. (2004) Image Retrieval with Visually Prominent Features using Fuzzy set theoretic Evaluation, ICVGIP 2004, India.
  5. Chen, Y. and Wang, J. Z. (2002) A Region-Based Fuzzy Feature Matching Approach to Content-Based Image Retrieval, IEEE Trans. on PAMI, Vol. 24, No.9, Pp. 1252-1267.
  6. Deselaers, T., Keysers, D. and Ney, H. (2008) Features for image retrieval: An experimental comparison, Information Retrieval, Vol. 11, Pp. 77-107.
  7. Gevers, T. and Smeuiders, A.W.M. (1999) Combining color and shape invariant features for image retrieval, Image and Vision computing, Vol.17, No. 7, Pp. 475-488.
  8. Goh, S.T. and Tan, K.L. (2000) MOSAIC: A fast multi-feature image retrieval system, Data & Knowledge Engineering, Vol. 33, Pp.219-239
  9. Ha, J., Kim, G. and Choi, H. (2008) The Content-Based Image Retrieval Method Using Multiple Features, Fourth International Conference on Networked Computing and Advanced Information Management, 2008. NCM '08, Pp. 652 – 657.
  10. Harris, C. and Stephens, M. (1988) A combined corner and edge detectors”, 4th Alvey Vision Conference, Pp. 147-151.
  11. Kokare, M. and Biswas, P.K. (2007) Texture image retrieval using rotated wavelet filters, Pattern Recognition Letters, Vol.28. Pp, 1240-1249.
  12. Li, J., Wang, J.Z. and Wiederhold, G. (2000) IRM: Integrated Region Matching for Image Retrieval, Proc. of the 8th ACM Int. Conf. on Multimedia, Pp. 147-156.
  13. Lin, H.J., Kao, Y.T., Yen, S.H. and Wang, C.J. (2004) A study of shape-based image retrieval, Proceedings of 24th International Conference on Distributed Computing Systems Workshops, Pp. 118 – 123.
  14. Manjunath, B.S. and Ma, W.Y (1996) Texture feature for browsing and retrieval of image data, IEEE Transaction on PAMI, Vol. 18, No. 8, Pp.837-842.
  15. Vadivel, A., Sural, S. and Majumdar, A.K. (2009) Image retrieval from the web using multiple features, Online Information Review, Vol. 33, Iss: 6, Pp.1169– 1188.
  16. Kass, M., Witkin, A. and Terzopoulos, D. (1988) Snakes: active contour models, Int. J. Comput. Vision, Vol. 9, Pp. 321-331.
  17. Lin, H.J., Kao, Y.T., Yen, S.H. and Wang, C.J. (2004) A study of shape-based image retrieval, Proceedings of 24th International Conference on Distributed Computing Systems Workshops, Pp. 118 – 123.
  18. Pakhira, M. K., Bandyopadhyay, S. and Maulik, U. (2004) Validity index for crisp and fuzzy clusters, Pattern Recognition, Vol. 37, Pp. 487-501.
  19. S.Rajasekaran, VasanthaKalyaniDavid , Pattern Recognition using Neural And Functional networks by 2008.
  20. R.Priya and Dr. VasanthaKalyani David (2010) Improved Content Based Image Retrieval Using Color Histogram And Self Organizing Maps, IJCSIS Vol 8, No. 9, Pp243-248
  21. R.Priya and Dr. VasanthaKalyani David (2011) Optimized Content based Image Retrieval System based on Multiple Feature Fusion Algorithm, IJCA Vol 31, No 8, Pp 33-37.