A Survey of Relevance Feedback Techniques in Content-Based Image Retrieval

Relevance feedback, as an effective approach to boost image retrieval, has become a necessary part of content-based image retrieval system, and attracted much research attention in the past few years. This paper provides a comprehensive survey of relevance feedback techniques described in the literature. After a brief introduction of content-based image retrieval, the interactive process of relevance feedback and its import aspects are discussed. Relevance feedback is further formulized as a supervised learning problem, and its characters are analyzed. Based on the retrieval model adopted in the algorithm, relevance feedback algorithms are categorized into three classes: distance-based approach, probabilistic approach, and machine learning based approach, and various representative algorithms are introduced following this categorization. At last, some promising research directions are also suggested.