The method proposal of image retrieval based on K-means algorithm

Abstract

In this paper, we propose a content-based image retrieval system using the improved K-means algorithm with binary indexes of images. The created index, known as binary signatures of image, is based on image features including shape, location, and color. Firstly, we present the method of creating binary signature based on CIE-L*a*b* color space and Discrete Wavelet Frames. After that, the similarity measure between two binary signatures is presented. On the basis of k-means algorithm, we propose several improvements for clustering binary signatures used later to assess similarities between images. From that, the clustering algorithm for binary signatures of images is proposed. Next, we give the image retrieval algorithm based on the partitioned signature clusters. For illustrating our theoretical proposal, some experiments are conducted on common image sets including COREL, CBIRimages, and WANG.