Abstract: Via the comparison of traditional histogram and segment based on histogram, a new image retrieval method of segment global histogram is proposed. The basic idea of segment global histogram is to divide image space into blocks according to a certain strategy, and then to calculate the color histogram of each block as the color feature of the corresponding block. Users assign a weight to each block according to the importance of the space information that the block contains. After the blocks with weight zero are merged into a single segment global histogram, on the system the distances between corresponding blocks with nonzero weight and the distance between partial segment global histograms are calculated. Then the overall distance accumulated according to weights given by users is regarded as the real distance between the two pictures. The pictures with less distances indicate less visional differences. Relevancefeedback is adopted to selfadjust the performance of the system.

Key words: image retrieval, feature extraction, color histogram