A Comprehensive Analysis for Relevance Feedback in CBIR System

X. Xie, Y. Zhao, and Z. Zhu (PRC)


Content-based image retrieval ( CBIR ) , relevance feedback, kernel density estimation, and mean shift


In order to narrow the gap between high-level concepts and low-level features, relevant feedback is a usually used technique in content-based image retrieval systems. In this paper, a comprehensive analysis for relevance feedback in CBIR system was conducted. By using a kernel function to estimate the distribution of query feature, a mean-shift based optimization technique is first adopted for query refining. To update the feature weight matrix, both the inter relations among feature vectors and intra relations among feature components are explored. Besides moving the query and updating the feature weight matrix, we also introduce a feature-database updating scheme to accumulate the useful semantic information. The final experimental results show that the proposed approach greatly improves the retrieval performance.

Important Links:

Go Back