Abstract
Many of color-based image retrieval methods use a single color histgram extracted from a whole image for similarity evaluation. However, these methods cannot represent spatial color information within the image. In this paper, to overcome the problem, we use multiple histgrams extracted from regions obtained by a segmentation of the image, and the relationship between the region and its adjacent regions for similarity evaluation between the images. Moreover, having an index for every region, our method can retrieve a part of the image. We show some results of image retrieval experiments by comparing our method with conventional one.
