Abstract

It is understood that the correlation process between similar patterns generally introduces ambiguity in decision making. This problem is much more profound in the binary pattern recognition process. For example, the cross-correlation between letters E and F exhibits the same main peak intensity as that in the autocorrelation of letter F. Many attempts have been made to overcome this difficulty, including the recognition technique based on a comparison of pattern’s perimeters, and the recognition of patterns in moment spaces or feature spaces. However, the phenomenon that causes the ambiguity can be converted and utilized as a feedback parameter when an adaptive process scheme is chosen in pattern recognition. In this paper, the relationship between pattern shapes and correlation results is analyzed first, followed by a proposed hybrid optical-electronic adaptive joint transform correlator. The adaptive capability of the system is achieved through interfacing between an optical correlator and a computer. The intensity distribution of correlation peaks detected in the optical correlator serves as a feedback to update reference images in the input plane of the correlator, so that an optimal decision can be made for the recognition process through adaptive iterations. In the iteration process, the saturating phenomenon in the intensity of correlation peaks is used as a guideline in designing a proper feedback scheme. Several uses in binary pattern recognition are demonstrated with results obtained from computer simulations as well as experimentally.

© 1991 Optical Society of America