FUZZY NEURAL-NETWORK FOR INVARIANT OPTICAL-PATTERN RECOGNITION

A novel fuzzy neural network (FNN) model for invariant pattern recogni tion is presented that combines fuzzy set reasoning and artificial neu ral network techniques. The presented FNN consists of three blocks: fu zzifier, fuzzy perceptron, and defuzzifier. It fuzzifies the input pat terns and trains the interconnection weights according to membership f unctions instead of traditional binary values. The proposed FNN has be en applied to 2-D binary-image pattern recognition under shift and som e other types of distortions. In comparison with the classical multila yer perceptron, the FNN possesses a higher recognition rate and is mor e robust to input distortions. (C) 1996 Society of Photo-Optical Instr umentation Engineers.