A Nonlinear Filtering Technique for Digitized Images Degraded by Film-Grain Noise,

摘要

A class of nonlinear recursive filtering and smoothing algorithms is investigated for use in digital processing of images degraded by film-grain noise. These algorithms are developed by using a Bayesian perturbation analysis and a class of 'almost Gaussian' conditional probability distributions to obtain results that are formally accurate to first order in the perturbation parameters. The nonlinear features of the resulting algorithms are used to account for the signal dependence of film-grain noise and the nonlinear relation between optical density and intensity. These algorithms are tested on a photographic film negative. Some improvement over conventional Wiener filtering is found. (Author)