An optimized learning strategy for image retrieval with relevance feedback

Relevance feedback for interactive image retrieval has been the focus of current research. In the process, users provide feedback in each iteration. The key issue in relevance feedback is how to effectively utilize the feedback information to improve retrieval performance. Neither using feedback information only from the current iteration nor using all feedback accumulated from the beginning is an optimized approach. To address this issue, we propose a feedback learning strategy, which adopts relevant images and adaptively classifies relevant images to construct optimal feedback information. By performing these two processes, knowledge accumulated from previous iterations is reasonably incorporated. The selected feedback information can capture the user´s intention and meet the user´s needs more precisely. Experimental results on collections of real-world images validate the efficacy of our proposed approach.