Weighted median cut quantization and its applications

This paper proposes an extended median cut quantization method with special median values, which does not have a restriction that the number of signals after quantization is a power of 2. The proposed quantization performs fast and the quantization error of it becomes smaller than that of the linear quantization for images with sparse histogram. When the number of the quantization levels is large, it is confirmed that the quantization error of the proposed quantization becomes smaller than the median cut quantization, and becomes almost the same as the Lloyd-Max quantization for general images. Finally, we apply to a high bit rate lossy data compression system for sparse histogram images using the proposed quantization.