An Integrated Approach to Shape Based Image Retrieval

Shape is one of the primary low level image features in the newly emerged Content Based Image Retrieval (CBIR). Many shape representations have been proposed, and they are generally classified into contour-based methods and region-based methods. Contour-based methods capture shape boundary features while ignore shape inner content. Region-based methods capture shape inner features while do not emphasize on boundary features. It is known neither methods can produce ideal retrieval results in all situations. The shortcoming of both types of methods can be overcome by combine the strength of two methods. In this paper, a contour shape descriptor and a region shape descriptor are studied. Then a integrated shape descriptor, which combines a contour shape descriptors and the region shape descriptors, is presented. The integrated shape descriptors provide interactivity between users and the retrieval system and can improve retrieval performance. Retrieval results are given to show the comparison and the improvement.