Color Image Adaptive Clustering Segmentation

Abstract

This paper presents an adaptive clustering segmentation approach based on fuzzy entropy and rival penalized competitive learning (RPCL) for color image. It can adoptively acquire appropriate number of color clusters and their centers of color image. Firstly, fuzzy entropy approach is applied to smooth color components' histograms and centers of each color component are determined. Then these centers of different color components are combined to form initial centers for RPCL. Finally, RPCL converges some of initial centers to actual centers of original color image and pushes the other initial centers away. The image is segmented by the former learned centers. The experiment shows that the method can effectively and adaptively segment the color images without specifying the number of initial clusters in advance.