Abstract

The Kittler-Young (KY) transform is a feature-extraction linear algorithm. The information of the class means and the class variances are optimally utilized in the KY transform, such that high space-bandwidth product pattern recognition can be performed. Another important feature of the KY transform is the capability to perform multiple-category pattern recognition. A three-category image classification is performed on an optical-matrix multiplication system by using the KY transform. The first three basis KY basis images, which are the column vectors of the KY transform matrix, are written onto the first liquid-crystal television (LCTV), while a target image is displayed on the second LCTV. After the optical inner-product and addition operations, the first three KY features of the target image are present at the output plane. These KY features are detected by a charge-coupled device camera and are fed to a pseudoinverse classifier.

© 1990 Optical Society of America