Content-based image retrieval by relevance feedback

Relevance feedback is a powerful technique for content-based image retrieval. Many parameter estimation approaches have been proposed for relevance feedback. However, most of them have only utilized information of the relevant retrieved images, and have given up, or have not made great use of information of the irrelevant retrieved images. This paper presents a novel approach to update the interweights of integrated probability function by using the information of both relevant and irrelevant retrieved images. Experimental results have shown the effectiveness and robustness of our proposed approach, especially in the situation of no relevant retrieved images.