Abstract
Accuracy and efficiency are two major issues in designing Content-Based Image Retrieval (CBIR) systems. Most of CBIR systems were dedicated to visual feature extraction because it has been shown that significant visual informations extracted from images can achieve similarity retrievals with high performance of effectiveness. However, in their studies, exhaustive searches were usually performed to generate a moderate number of plausible retrievals in response to a user's query. As a result, retrieval efficiency becomes a bottleneck when the matching process is progressed over a large volume of image