Image retrieval using local color features

In this study, an image indexing and retrieval approach using local color features and a modified weighted color distortion measure is proposed. In the proposed approach, each image is segmented into several regions by a watershed segmentation algorithm, and then the mutual relationships between connected color regions are extracted as local color features. That is, an image can be represented as a set of connected (adjacent) color regions and the mutual relationships between connected color regions. In the image retrieval stage, the similarity between a query image and a target image will contain not only direct region correspondence but also the mutual relationships between connected color regions. A modified weighted color distortion measure is proposed, in which different color elements in the YUV color space will receive different weights so that the illumination variation effect will be greatly reduced.