A CBIR CLASSIFICATION USING SUPPORT VECTOR MACHINES

Content Based Image Retrieval (CBIR) is a developing trend in Digital Image Processing for searching and retrieving the query image from wide range of databases. Conventional content-based image retrieval (CBIR) schemes have following limitations: 1. It is slow 2. difficult to label negative examples; 3. Accuracy is poor in a single step; 4. users may introduce some noisy examples into the query. This inturn explores solutions to a new issue that image retrieval using unclean positive examples. This paper proposes a new two-step strategy in which first step is feature extraction using low level features (color, shape and texture) while SVM classifier is used in the second step to handle the noisy positive examples. Thus, an efficient image retrieval algorithm based on color-correlogram for color feature extraction, wavelet transformation for extracting shape features and Gabor wavelet for texture feature extractionis proposed. Further, multiple features and different distance metrics are combined to obtain image similarity using SVM classifier. Results based on this approach are found encouraging in terms of color, shape and texture image classification accuracy. After the features are selected, an SVM classifier[4] is trained to distinguish between relevant and irrelevant images accordingly.