Abstract

A trinary (or ternary) neuron representation for Hopfield's associative memory¹ has been recently found to be more effective in associative recall. The trinary associative memory overcomes the discrepancies associated with the unipolar binary and bipolar binary representations of neurons.² In the trinary neuron representation, the known portion of the partial input is represented in bipolar binary and the unknown portion is represented by a string of 0s. In this work, the trinary associative memory is used to process 1-D as well as 2-D stored images such as those encountered in character recognition applications.

© 1989 Optical Society of America