Distance selection based on relevance feedback in the context of CBIR using the SFS meta-heuristic with one round

In this paper, we address the selection in the context of Content Based-Image Retrieval (CBIR). Instead of addressing features' selection issue, we deal here with distance selection as a novel paradigm poorly addressed within CBIR field. Whereas distance concept is a very precise and sharp mathematical tool, we extend the study to weak distances: Similarity, quasi-distance, and divergence. Therefore, as many as eighteen (18) such measures as considered: distances: {Euclidian, …}, similarities{Ruzika, …}, quasi-distances: {Neyman-X2, …} and divergences: {Jeffrey, …}. We specifically propose a hybrid system based on the Sequential Forward Selector (SFS) meta-heuristic with one round and relevance feedback. The experiments conducted on the Wang database (Corel-1K) using color moments as a signature show that our system yields promising results in terms of effectiveness.