A Relevance Feedback Method in Image Retrieval with Analyzing Feedback Log

Content-based image retrieval is one of the most important research areas in multimedia database. In recent years it has been more and more popular to use the mechanism of relevance feedback to improve retrieval performance. However, due to the interactive and iterative characteristics of the relevance feedback, the efficiency is still the main obstacle of the application of this technology in practice. To address this problem a novel approach is presented, which accelerates feedback process by analyzing feedback log file using collaborative filtering model. Through analyzing the log file, the system may get some correlation information to predicate the images’ semantic similarity with respect to the current retrieval. A prototype image retrieval system is implemented, and the experiment result show that compared with the traditional relevance feedback method, the retrieval performance can be improved apparently.