RELEVANCE FEEDBACK - A POWER TOOL FOR INTERACTIVE CONTENT-BASED IMAGERETRIEVAL

Content-based image retrieval (CBIR) has become one of the most active research areas in the past few years.. Many visual feature representa tions have been explored and many systems built. While these research efforts establish the basis of CBIR, the usefulness of the proposed ap proaches is limited. Specifically, these efforts have relatively ignor ed two distinct characteristics of CBIR systems: 1) the gap between hi gh-level concepts and low-level features, and 2) subjectivity of human perception of visual content. This paper proposes a relevance feedbac k based interactive retrieval approach, which effectively takes into a ccount the above two characteristics in CBIR. During the retrieval pro cess, the user's high-level query and perception subjectivity are capt ured by dynamically updated weights based on the user's feedback. The experimental results over more than 70000 images show that the propose d approach greatly reduces the user's effort of composing a query, and captures the user's information need more precisely.