Abstract: In this paper, a modified Paik-Hopfield neural network model based on the new state updating rule is proposed for restoring a degraded image.The convergence, the residual error and the energy change of the full parallel mode are thoroughly studied.An improved iteration algorithm based on the idea of "from coarse to fine" and the difference estimation between two adjacent layers is also presented.Experimental results demonstrate that this method can approximate the minimum of the energy infinitely and greatly improve the speed of convergence as well as the precision.

Key words: Hopfield neural network, regularization, image restoration, full parallel algorithm