A consensus-based fusion algorithm in shape-based image retrieval

Abstract

Shape-based image retrieval techniques are among the most successful content-based image retrieval (CBIR) approaches. In recent years, the number of shape-based image retrieval techniques has dramatically increased; however, each technique has both advantages and shortcomings. This paper proposes a consensus-based fusion algorithm to integrate several shape-based image retrieval techniques so as to enhance the performance of the image retrieval process. In this algorithm, several techniques work as a team: they exchange their ranking information based on pair-wise co-ranking to reach a consensus that will improve their final ranking decisions. Although the proposed algorithm handles any number of CBIR techniques, only three common techniques are used to demonstrate its effectiveness. Several experiments were conducted on the widely used MPEG-7 database. The results indicate that the proposed fusion algorithm significantly improves the retrieval process.