Skip to Main Content
This paper presents a novel BP-based image retrieval (BPBIR) system, which is based on the observation that the images users need are often similar to a set of images with the same conception instead of one query image and the assumption that there is a nonlinear relationship between different features. If users aren't satisfied with the retrieved results, relevance feedback method is used to enhance the performance of the proposed system by changing the weights of the BP neural networks. In addition, we discuss some divisional methods to give rough information on the spatial color composition. Finally, we compare the performance of the proposed system with other systems. Experimental results show the efficacy of the proposed system.