Non-causal error diffusion for image halftoning

In conventional error diffusion (ED) for image halftoning, a causal filter is commonly used to diffuse the error signal. The causality makes the ED algorithm very simple and easy to implement, but also leaves many undesirable patterns in halftones. In this paper, we proposed to adopt non-causal ED and to model the non-causal ED as the sampling of an appropriate Markov random field. Because of the MRF modeling, the non-causal ED is feasible by using conventional Gibbs sampler. Furthermore, due to the non-causality, the halftones obtained were clearer, without periodic or directional patterns, and with more image details. This can be seen from the experimental results consisting of the gray-level ramp image and nature images.