Probabilistic Modeling for Inverse Halftoning using Edge-Preserving Prior

Y. Saika and K. Okamoto (Japan)

Keywords

Bayesian inference, inverse halftoning, the edge preserving prior, Monte Carlo simulation

Abstract

On the basis of the Bayesian inference using the maximizer of the posterior marginal (MPM) estimate, we construct the probabilistic modeling to the problem of inverse halftoning for a halftone version of a gray-level image which is obtained by the dither method. We here try a model of the edge-preserving prior to the MPM estimate in the hope that the present method effectively reconstructs both smooth structures and edges which appear between smooth structures. Then, in order to clarify the performance of the present method, we first carry out the Monte Carlo simulation for a set of gray level images generated by the edge-preserving true prior which has the same form as the assumed model prior. The Monte Carlo simulation reveals the results that the present method achieves the optimal performance around the Bayes-optimal condition and further that the Bayes optimal solution works more effectively than the MAP estimate which uses the same cost function as the MPM estimate. Next, using the Monte Carlo simulation for the 256-level standard image “Lena” we also obtain the result that the present method works as well as the MAP estimate even for the realistic gray-level image.

Important Links:



Go Back