Relevance Feedback for Distributed Content Based Image Retrieval

This paper investigates a decentralized content-based image search system with a distributed. At the end of the indexing phase, the feature-descriptors are partitioned into multiple clusters using self-organizing tree map. At the retrieval phase, feature-descriptors of a query image are firstly used to short-list clusters for performing search operation, and relevance feedback using radial basis functional network is adopted for improving the retrieval precision. The study in this paper demonstrates that pre-processing and decentralizing feature descriptor database successfully help reducing and offloading computational demand, and the proposed relevance feedback approach for the distributed CBIR system helps improving the retrieval precisions.