Color classification of natural color images

Abstract

The present article describes a color classification method that partitions a color image into a set of uniform color regions. The input image data are first mapped from device coordinates into the CIE L*a*b* color space, an approximately uniform perceptual color space. Colors used to represent a natural color image are classified by means of cluster detection in the uniform color space. The basic process of color classification is based on histogram analysis to detect color clusters sequentially. The principal components of a color distribution are extracted for effective discrimination of clusters. We present an algorithm for sequential detection of color clusters in the uniform color space, and the related algorithms for region processing and color computation. The performance of the method is discussed in an experiment using three kinds of natural color images.