Image retrieval based on color coherence

The color histogram is a simple method for representing and comparing images, but this approach has a drawback that it does not capture spatial image attributes. Previously, color coherence vectors based on color homogeneous regions have been introduced for image retrieval since they include some spatial information. The color histogram constructed by connected components whose size is larger than a certain number of pixels. This definition takes only the spatial size into consideration. In this paper, we present new definitions of color coherence that can include spatial information, say, distance between regions. As a result, we have three types of color coherence measures. Pixels are grouped according to (1) region size only, (2) distance only and (3) both size and distance. We apply them to actual visual image retrieval applications. For the simulation, we proposed a fast segmentation and merging method. The experimental result showed that the distance information is more effective than the size information for the image retrieval.