IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.4, April 2008 327 Manuscript received April 5, 2008 Manuscript revised April 20, 2008 Content Based Image Retrieval Using Independent Component Analysis Arti Khaparde † B.L.Deekshatulu †† M.Madhavilatha ††† Zakira Farheen † Sandhya Kumari V † †Department of Electronics and Communication, Aurora’s Technological and Research Institute, Hyderabad, India †† Department of Computer Science and Artificial Intelligence, University of Hyderabad, Hyderabad, India. ††† Department of Electronics and Communication, Jawaharlal Nehru Technological University, Hyderabad, India. Summary Content Based Image Retrieval (CBIR) has become one of the most active research areas in the past few years. Many indexing techniques are based on global features distribution such as Gabor Wavelets. [1]. In this paper we present a new approach for global feature extraction using an emerging technique known as Independent Component Analysis (ICA). A comparative study is done between ICA feature vectors and Gabor feature vectors for 180 different texture and natural images in a databank. Result analysis show that extracting color and texture information by ICA provides significantly improved results in terms of retrieval accuracy, computational complexity and storage space of feature vectors as compared to Gabor approaches. Key words: CBIR, Gabor Wavelets, ICA, Similarity measurements 1. Introduction Recent years have witnessed a rapid increase of the volume of digital image collection, which motivates the research of CBIR. To avoid manual annotation many alternative approaches were introduced by which images would be indexed by their visual contents such as color, texture, shape etc. Many research efforts have been made to extract these low level image features, evaluate distance metrics and look for efficient searching schemes. A CBIR is a two step approach to search the image in database. First, for each image in the database, a feature vector is computed and stored in feature database. Second given a query image, its feature vector is compared to the feature vectors in the data base and images most similar to the query image are returned to the user. The feature and similarity measure used to compare two feature vectors should be efficient enough to match similar images. We have presented ICA of images as a computational technique for creating a new data dependent filter bank. The new ICA filter bank is similar to the Gabor filter bank but it seems to be richer in the sense that some filters have more complex frequency responses. They are able to capture the inherent properties of textured images. The ICA based approached is different from existing filtering methods in that it produces a data dependent filter bank.[6] This paper describes an image retrieval technique based on ICA and the results are compared with the Gabor features. We demonstrate our retrieval results both for texture images and for natural images. The paper is organized as follows: Section 2 describes fundamentals of 2-D Gabor filters. Section 3 describes ICA. Section 4 discusses similarity measurement techniques used for retrieval. In section 5, we present experimental results of image retrieval based on Gabor as well as ICA feature vector. Section 6 concludes the paper. 2. Gabor Filter Wavelets Gabor wavelet is widely adopted to extract texture features from the images for retrieval and has been shown to be very efficient [9,11]. Basically Gabor filters are a group of wavelets, with each wavelet capturing energy at a specific frequency and specific orientation. The scale and orientation tunable property of Gabor filter makes its especially useful for texture analysis. The design of Gabor filter is done as follows: Gabor Filter (wavelet)[8] For a given image I(x,y) with size PXQ, its discrete Gabor wavelet transform is given by a convolution: * (,) ( , ) (,) mn mn s t G x y Ix s y t st ψ = − − ∑∑ (1) where, s and t are the filter mask size variables, and * mn ψ is a complex conjugate of mn ψ which is a class of