IMAGE RESTORATION USING RANDOM FIELD MODELS.

The use of two classes of two-dimensional Random Field (RF) models, known as the simultaneous and conditional Markov RF models, to develop algorithms for the minimum mean squared error (MMSE) restoration of degraded images is considered. The degradation is assumed to be due to a space invariant, periodic, nonseparable point spread function and an additive white Gaussian noise. The restoration algorithm is optimal with respect to MMSE criterion if the parameters characterizing the RF models are exactly known. However, in practice the parameters are estimated from the original image. An iterative scheme is presented for the estimation of unknown parameters. The restoration algorithm has a general structure that includes various types of RF models, such as casual, semicausal and noncausal models. Examples of restoration are given.