Abstract

Both the intensity of correlation peaks and the correlation distribution¹ have been utilized in a binary pattern recognition process to obtain information on the contour of the input object. In this paper, the symmetry properties of the correlation distribution are investigated and introduced as a parameter in classifying input objects. An example of identifying alphabetical letters from this parameter is presented. A horizontal bar and a vertical bar are chosen as selected features in the correlation process. Based on the symmetric properties of the correlation distribution, different categories can be classified. Further identification of individual letters in the same class can be achieved by the comparison of correlation peak intensities. The proposed approach drastically reduces the number of references required, compared to some conventional filtering methods, in identifying input objects from a given set of images. In addition, the technique presented shows promising performance on noise tolerance and scale invariance.

© 1992 Optical Society of America