Abstract

Markov random fields (MRFs) [1, 2, 3, 4] provide attractive statistical models for multidimensional signals. However, unfortunately, optimal Bayesian estimators tend to require large amounts of computation. We present an approximation to a particular Bayesian estimator which requires much reduced computation and an example illustrating low-light unknown-blur imaging. See [7] for an alternative approximation based on approximating the MRF lattice by a system of trees and for an alternative cost function.

© 1995 Optical Society of America