A learnable cellular neural network structure with ratio memory for image processing

Abstract

In this paper, a learnable cellular neural network (CNN) with space-variant templates and ratio memory (RM) called the RMCNN, is proposed and analyzed. By incorporating both a modified Hebbian learning rule and RM into the CNN architecture, the RMCNN as an associative memory can generate the absolute weights and then transform them into the ratioed A-template weights as the ratio memories for recognition of noisy input patterns. It is found from simulation results that, due to the feature enhancement effect of RM, the RMCNN under constant leakage on template coefficients can store and recognize more patterns than CNN associative memories without RM, but with the same learning rule and the same constant leakage on space-variant template coefficients. For 9/spl times/9 (18/spl times/18) RMCNNs, three (five) patterns can be learned, stored and recognized. Based upon the RMCNN architecture, an experimental CMOS 9/spl times/9 RMCNN chip is designed and fabricated by using 0.35 /spl mu/m CMOS technology. The measurement results have successfully verified the correct functions of RMCNN.