A fast-converging Hamming net used in an offline Chinese character recognition system

The authors (1991) previously proposed a revised version of a holographic memory model based on adaptive feature detection with an attention shift switch. To implement it in a Chinese character recognition system, a fast-converging Hamming net with two memory layers is proposed corresponding to two shiftable stages of feature extraction in the Chinese recognition system. The attention shift process is realized automatically. The system was used to learn 50 Chinese characters. With a recognition test on a set of 30 samples for each character, a recognition rate of about 85% and a recognition speed of about 3.3 words per second were achieved. >