An Expert Technique for Content-Based Color Image Retrieval

Abstract

Recently, as web and various databases contain a large number of images, CBIR (content-based image retrieval) are greatly used. This paper proposes an image retrieval system using color-spatial information from the applications. First, we suggest two kinds of the indexing keys to prune away the irrelevant images to given query images using MCS (Major Color Set) and DBS (Distribution Block Signature). MCS's are related to color information while, DBS's are related to spatial information respectively. After successively applying these filters to a large database, we get only a small amount of high potential candidates that is somewhat similar to that of query images. We propose to use QM (quad modeling) method to set the initial weight of 2-dimensional cell in the query image according to each major color and retrieve more similar images through similarity association function associated with the weights. Finally, we evaluated the system's efficiency by statistically how many images were expected to be filtered out during the first and second filtering processes.