Content-based image retrieval system with new low-level features, new similarity metric, and novel feedback learning

A currently relevant research field in computer science is the management of multimedia databases. Two related key issues are achieving an efficient content-based retrieval and a fast response time. Relevance feedback is a powerful tool to improve the retrieval results of the CBIR systems. However the traditional relevance feedback could only search a small feature subspace comparing with the entire huge feature space. So, this paper provides solutions to enlarge the searching feature subspace in a CBIR system, effectively. Firstly, an adaptive system query strategy to user behaviors is introduced to improve the performance of relevance feedback. Then a feedback scheme based on multi-features classification is developed. Both tactics enlarge the searching feature subspace efficiently. From experimental results, it is clear that the feedback scheme has a much better performance than the traditional CBIR systems. To develop such a scheme, a new effective texture feature and an efficient way to measure the dis-similarity between two image features are proposed in the CBIR system, which provides solutions to a general query on a large image database with 56,600 images.