Abstract

The Kittler-Young (K-Y) transform is a nonparametric method for feature selection, which does not require detailed assumptions about the probability structure of the problem and has obvious practical potential. The feature selection criterion for K-Y transform is the averaged mutual distance between class means. Thus, the K-Y transform maps random input images into a feature space in which the differences between class means are maximized. And the information on the class means and variances is utilized optimally. Using a joint transform correlator (JTC), the K-Y transform has been implemented optically to solve a two-class problem of image classification. Having calculated the K-Y basis images, which are real in general, the two most significant K-Y basis images are directly written onto a LCTV at the input plane of a JTC as reference images with a bias added. The outputs from the JTC, which take the form of bright spots, are the correlations between the input image to be classified and the reference images. Finally, these bright spots are detected electronically and the feature vector is taken into a microcomputer to perform a classification.

© 1989 Optical Society of America