A multiscale color error diffusion algorithm for color quantization

This paper presents a new multiscale vector error diffusion algorithm for color quantization. This iterative algorithm selects a pixel, color-quantizes it and then diffuses the quantization error with a non-causal filter at each iteration until all pixels are color-quantized. By doing so, directional hysteresis is completely eliminated and color impulses can be reduced to a certain extent. Simulation results show that it provides better outputs in terms of S-CIELAB color difference metric as compared with conventional approaches. The subjective quality of the output is also improved.