Blind bilevel image restoration using Hopfield neural networks

Abstract

A Hopfield neural network approach to blind bilevel image restoration is presented. In the approach, two kinds of Hopfield neural networks are used. One is the analog Hopfield neural network, utilized to estimate the parameters of the finite point spread function (PSF) of a blurring system. The other one is the modified Hopfield neural network used to restore bilevel image. The entire model is based on the alternative operation of the two networks. In the modified Hopfield neural network, the eliminating highest error (EHE) criterion is applied for the purpose of obtaining a more precise solution. Simulation results show that, after a few iterations, the model always obtains a bilevel image whose quality is almost the same as, or even better than, what is obtained by the modified Hopfield network when the precise parameters of PSF are used. The results are quite good. If the EHF criterion is not used, the model does not achieve a good bi-level image. >