A classification approach for estimating prior models in statistical image restoration

We present a novel application of Bayesian classifier to the problem of statistical image restoration which has attracted much attention in recent years. Specifically, maximum a posteriori (MAP) based approaches have been applied fairly successfully to a variety of image data. An integral part of such restoration schemes is the estimation of the prior probability of occurrence of an estimate of the ideal image. Generally speaking, this is employed as a means of reducing noise in the restored image. We present a new approach with the aim of modelling the local characteristics of ideal images based on the use of a set of training data. This has led to the development of a versatile and robust Bayesian restoration method for imaging applications.