In this paper, we propose an improved color quantization method. It compensates a disadvantage of conventional k-means color quantization. Usually, color quantization has number of colors to quantize image. However, because of k-means color quantization initializes its centroids of clusters with random values, it cannot use whole colors. The proposed method uses popularity color quantization to initialize the centroids of clusters. To evaluate the proposed method, we use ten images. The experimental results show the proposed method improved the conventional k-means color quantization. It has average 25.11% less of mean absolute error than conventional k-means color quantization environment.

color quantization, popularity, k-means, image processing.