Image restoration using neural networks

Two neural algorithms for image restoration are proposed. The image is considered degraded by linear blur and additive white Gaussian noise. Maximum a posteriori estimation and regularization theory applied to this problem lead to the same high dimension optimization problem. The developed schemes, one having a sequential updating schedule and the other being fully parallel, implement iterative minimization algorithms which are proved to converge. The robustness of these algorithms with respect to finite numerical precision is studied. Examples with real images are presented.<>