A context-based fusion algorithm for shape retrieval

Shape-based image retrieval has demonstrated encouraging results in retrieving images based on their content. A large body of research in this area has focused on finding effective shape descriptors to search for query images. Nevertheless, the boundless image content variation makes it impossible for a particular choice of descriptor and an algorithm to be effective for all types of images. It is therefore reasonable to approach the problem by combining a group of descriptors and algorithms. Since the effectiveness of an algorithm/descriptor is image dependent, we maintain that for this strategy to achieve its intended goal, the combination scheme must be dynamic as a function of the query context. This paper proposes a context-based fusion algorithm to integrate several shape-based retrieval techniques. Query context is used to determine the most appropriate technique or combination that yields optimal performance. In testing the proposed algorithm, several experiments were conducted.