RWM-cut for color image quantization

Abstract

A new simple method for constructing a color palette that uses the radius weighted mean cut (RWM-cut) is proposed. The method is a hierarchically divisive method, and each two-class partition uses the centroid and the RWM only. Experiments show that the RWM-cut algorithm is feasible and visually acceptable. The algorithm can either be used alone or be used to create a good initial palette for the LBG algorithm. Besides the 3-D version, a 1-D version of the RWM-cut algorithm is also included in the paper for realtime color quantization. The quantization error is small and the processing speed is competitive. Dithered images are also provided. Copyright (C) 1996 Elsevier Science Ltd