A color-based retrieval system for images collected on the World Wide Web

This work presents a prototype system for color-based retrieval of images collected on the Web. The images collected are classified in two semantic classes (photographs and graphics) but only photographs are indexed. It basically stores information about regions with the same color in the image (chromatic regions), such as size, position and boundaries with other regions. This data representation is made using histogram variations of the image as graphs: the modified color adjacency graph (MCAG) and the variance graph (SVG). Image retrieval is done by applying a similar metric based on a comparison of the graphs.