Performance of generalized statistical smoothing to inverse halftoning

We construct a method of inverse halftoning for a halftone version of a grayscale image converted by the error diffusion method via the Floyd-Steinberg kernel by making use of the generalized statistical smoothing which is constructed by introducing both edge enhancement procedure and generalized parameter scheduling into the statistical smoothing originally proposed by Wong. Then, in order to clarify the performance of the present method, we numerically estimate the mean square error and the mean square error between original and reconstructed images modulated by the MTF function of the human vision system. Using numerical simulations for several 256-level standard images, we clarify that the optimal performance is realized by introducing the edge enhancement and the generalized parameter scheduling, if we tune parameters appropriately. Then, we find that the present method reconstructs original images with high image quality, if we introduce the appropriate models both for the edge enhancement procedure and the generalized parameter scheduling.