We present a new algorithm for color image quantization based on human color perception properties. We construct two kinds of map by analyzing the spatial color distributions to take account of the human visual system; homogeneity map (H-map) and distinctiveness map (D-map). Then, we assign weight value to all color vectors by combining these maps to consider two factors at the same time. To extract representative colors, we define a new cost function and use the LKMA (local KMeans algorithm) with weighted color vectors. In this stage, we utilize an incremental splitting scheme with a penalty term to determine optimal number of clusters adaptively. The experimental results show that proposed algorithm reproduces an image preserving significant local features while removing unimportant details of an original image from the viewpoint of human.