Color quantization using modified median cut

We describe some observations on the practical implementation of the median cut color quantization algorithm, suitably modified for accurate col or rendering. The RGB color space is successively divided in such a way that colors with visual significance, even if relatively small in population, are given representatives in the color map. Appropriately modified, me- dian cut quantization is nearly as good as our best octree quantization. As with octree quan- tization, error-diffusion dithering is useful for reducin g posterization in large regions with a slow variation in color.