A novel image retrieval using color quantization and indexing histogram

Huazhang Wang,Xiaohai He

2007-01-01

Journal of Computational Information Systems

Abstract:A novel content-based image retrieval (CBIR) algorithm using color quantization and indexing histogram is proposed. The scheme is integrated with human vision. First, the image is divided into 4 by 4 non-overlapping blocks. According to different contribution to human vision of image blocks, each block is classified as uniform or non-uniform based on its gradient magnitude. In RGB space, we use the average of each color component for the uniform block. For each non-uniform block, we find out two representative colors to assign the non-uniform block by color moment-preserving. Then we construct the corresponding indexing histogram of uniform and non-uniform blocks. Finally, we search similar images by integrated features. Experimental results demonstrate that the retrieval performance of the proposed scheme is more accurate than that of other similar methods.