Optical pattern recognition by extracting least substructuring elements

Optical pattern recognition and classification are commonly implemented by means of correlation, and a specialized filter for the patterns to be recog nized must be constructed first. Although it has the inherent advantages of high speed, shift invariance, and high distinguishability, how to build an effective and easily implemented filter or group of filters still remains an open problem. By combining the methods of correlation filtering, transfo rm encoding, and neural network mapping, a least substructuring elements (L SE) extracting method is proposed in this paper. Some basic substructuring elements of a specific group of patterns to be processed could be extracted to compose masks in the least number. Computer simulation upon the 26 Engl ish capital letters is provided. One integrated hybrid optoelectronic imple mentation system is described, (C) 1999 Society of Photo-Optical Instrument ation Engineers. [S0091-3286(99)00710-2].