Non-separable weighted median-cut quantization for images with sparse color histogram

Non-separable color reduction for color images is proposed in this paper. The proposed method befits high bit rate quantization in which, for example, a 24 bits per pixel (bpp) color image is quantized to an image with 18 bpp or more, while the conventional median-cut quantization gives huge quantization errors in such condition. This feature is useful in bit depth conversion for modern high bit depth images and is also useful in visually lossless compression. Moreover, by taking into account color histogram sparseness in which a few colors among huge possible colors are used, lossless quantization with a reasonable processing time is served by the proposed method, whereas the conventional vector quantization requires numerous processing time for lossless quantization.