Semantic Clustering for Region-Based Image Retrieval

This paper proposes a clustering scheme to reduce search space and gap, two most challenging tasks in content-based image retrieval. By performing clustering before image retrieval, the search space can be reduced to those clusters that are close to the query target. In the proposed method, image sub-regions/segments are grouped into clusters in terms of their meanings in addition to their low level features. Ideally, one cluster approximates one concept or a small set of closely related concepts; hence the semantic gap in the retrieval phase is reduced. The experimental results show the effectiveness of the proposed method.