基于颜色分块全局直方图的图像检索方法及系统实现

Image Retrieval Methods and System Implementation Based on Color Segment Global Histograms

通过比较全局颜色直方图和分块颜色直方图,提出一种分块全局直方图的检索方法.该方法以分块直方图的特征为基础,在检索时,由用户为含有重要空间信息的分块设置权值,并将不注重空间信息的分块权值设为零.系统将权值为零的分块重新合并为部分全局直方图,计算所有权值非零分块的距离及部分全局直方图的距离,再按照用户给出的权值进行加权累加作为两幅图像之间的最终距离.结果表明,两幅图像间的距离越小,则视觉差异也越小.同时,引入相关反馈技术以改善检索效果.

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.