Performance of generalized statistical smoothing to inverse halftoning due to human vision system

We constructed a method of inverse halftoning using generalized statistical smoothing (GSS) for a halftone image obtained by converting a grayscale image with the use of error diffusion method via the Floyd-Steinberg kernel. This method was constructed by introducing edge enhancement and generalized parameter scheduling into a method of statistical smoothing (SS) proposed by Wong. In order to clarify the performance of the GSS, we evaluated both the mean square error and the mean square error modulated by the MTF function representing the sensitivity of human vision system. Numerical simulations for 256-grayscale standard images showed that optimal performance is achieved by introducing the appropriate models of edge enhancement and generalized parameter scheduling. Also, we found that the optimal performance is superior to those of the SS and the conventional Gaussian filter.