Restoration of degraded character dot image using discrete Hopfield neural network

Abstract

In order to estimate the image restoration capability of discrete type Hopfield neural network having an associative memory effect, some simulation experiments were performed. The memorized patterns to the network are binary dot alphabet capital characters consisting of 10/spl times/10 pixels. On the other hand, artificially degraded binary dot patterns of those original characters are used as the input patterns for the neural network to recall the original characters. As a result, the rate of success to recall the correct pattern is strongly related to both degradation degree of input patterns and number of patterns previously memorized in the network.