Spatial homogeneity-based fuzzy c-means algorithm for image segmentation

Abstract

A fuzzy c-means algorithm incorporating the notion of dominant colors and spatial homogeneity is proposed for the color clustering problem. The proposed algorithm extracts the most vivid and distinguishable colors, referred to as the dominant colors, and then used these colors as the initial centroids in the clustering calculations. This is achieved by introducing reference colors and defining a fuzzy membership model between a color point and each reference color. The objective function of the proposed algorithm incorporates the spatial homogeneity, which reflects the uniformity of a region. The homogeneity is quantified in terms of the variance and discontinuity of the spatial neighborhood around a color point. The effectiveness and reliability of the proposed method is demonstrated through various color clustering examples.