Feature filtering in relevance feedback of image retrieval based on a statistical approach

Abstract

Relevance feedback is a powerful tool to grasp the user's intention in image retrieval systems and has attracted many researchers' attention since 90's. In this paper, a feature filter whose parameters are computed by a statistical re-sampling approach is proposed in order to select the unique features to characterize the positive samples. A statistical voting procedure is then adopted to rank the candidates after getting rid of irrelevant feature components. Experimental results show that the proposed approach is more efficient and robust than the traditional method.