Probabilistic Error Diffusion For Image Enhancement

This paper presents a stochastic error diffusion approach and analyzes its effect on image quality and artifacts. Unlike traditional error diffusion techniques that use fixed or image- dependent error diffusion weights, stochastic error diffusion distributes image quantization errors to neighboring pixels based on a stochastic process whose expectation and entropy are controllable. A model is provided to derive diffusion weight expectations based on the principle of "uniform error propagation ". The paper empirically compares the difference between random quantization thresholds and random diffusion weights and shows that stochastic error diffusion, unlike random dithering, does not generally produce images of degraded quality. On the contrary, by selecting appropriate expectation and entropy for diffusion weights, stochastic error diffusion may drastically reduce image noise patterns with little loss of image sharpness.