An eliminating highest error criterion in Hopfield neural network for bilevel image restoration

In this approach to bilevel image restoration the autoconnections of the network generally weight more heavily than interconnections. This characteristic exists in general degradation models of image restoration and can be utilized to guide the network to be updated more efficiently. A criterion for choice of the neurons to be updated at each step is proposed. An algorithm using the criterion converges to more precise solutions with fewer updates as shown by simulation