A Feature Relevance Estimation Method for Content-Based Image Retrieval

Feature relevance estimation is one of the most successful techniques used for improving the retrieval results of a content-based image retrieval (CBIR) system based on users’ feedbacks. In this class of approaches, the weights of the feature elements (FEs) are adjusted based on the relevance feedbacks (RFs) given by the users to reduce the socalled semantic gap in the underlying CBIR system. An analytical approach is proposed in this paper to convert the users’ feedbacks to the appropriate FE weights by solving a constrained optimization problem. Experiments on a set of 11,000 images from the Corel database show that the proposed approach outperforms other existing short-term RF approaches reported in the literature. The proposed approach is also incorporated in two long-term RF methods and enhanced their performance.