This paper presents a novel technique for content based image retrieval (CBIR) that selects and assigns weights to the regions of the image on the basis of their contribution to image contents, using a new region-weight assignment scheme. Assigning the weight to each region ignores the irrelevant regions of the image during retrieval and thus maximizes the retrieval accuracy. The proposed approach performs the feature extraction at both region-level and image-level. Texture and edge features are extracted at region-level whereas shape feature is extracted at image-level. At region-level, the image is divided into non-overlapping regions and texture and edge features are calculated for each region separately. Curvelet transform is used for extracting the texture feature using the curve continuity as well as line continuity in the feature extraction process. Moment invariant is used for extracting the shape features. Integrated Region Matching (IRM) technique is used for retrieving the relevant images. The proposed approach does the best use of the features by balancing the regions and features in the similarity matching of the regions. The performance of the proposed technique is tested on COREL and CIFAR databases. Experimental results show the effectiveness of proposed region weight assignment scheme over the feature weight assignment scheme in image retrieval in comparison to other state-of-the-art techniques.
|