Analysis for identifying character with noise or defect via discrete Hopfield neural networks

Discrete Hopfield neural networks have memory and association function. In this paper, we use Hopfield neural networks to identify characters with noise or defect. We first make 26 English letters become network attractors through appropriate coding and orthogonalization processing, then study according to the Hebb learning rule. When input a noisy or defective letter, it will converge to the correct letter according to state evolution equation of the network. The orthogonalization method can also improve memory capacity of the network in the same network scale. And by MATLAB simulation, We can illustrate the validity and feasibility of the method.