Relevance feedback techniques in the MARS image retrieval system

Abstract

Content-based image retrieval (CBIR) has become one of the most active research areas in the past few years. Many visual feature representations have been explored and many systems built. In this paper, we focus on an important component of these systems - relevance feedback - and how we incorporated it into the MARS retrieval system. Relevance feedback techniques are based on an interactive retrieval approach to effectively take into account user preferences to provide an improved search experience. We present a series of coherent strategies, from single-point to multipoint and multi-feature approaches that we have seamlessly integrated into our system and present experimental results to show their retrieval performance characteristics.