Optical multiple-correlation system for pattern recognition

Abstract

We have developed an optical multiple correlation system for pattern recognition. It has superior ability of pattern discrimination even though an input image contains an unknown and complicated background or noise, which often disturbs the pattern recognition in the real world application. A design method for the multiple correlation filters has been developed, in which correlation filters are synthesized with component images extracted from training images by the principal component analysis. It is shown the correlation filters designed by this method has better efficiency on the noise tolerance and the detection ability than the filters synthesized directly with all training images. An optical multiple-correlation system has been developed and it is applied to a vision system of the robot named 'Gazing Tiger', in which the optical multiple correlation system detects locations of human faces in crowds and the robot looks a human face producing largest signal of correlation. The developed optical system can operate well even in the crowd and in the complicated background. It denotes the superior ability of the multiple correlation system we have developed.