Abstract

The Karhunen-Loeve (K-L) transform has been applied to the training of optical neural networks. Because orthogonal images are used in the training, the K-L training process improves the Hopfield-model capacity of a neural network. Moreover, pattern recognition based on the structure of images and the Hamming distance is achieved for neural networks. For the design of optical neural networks, the relation between the minimum focal length of the two-dimensional optical neural networks and the maximum number of neurons is derived.

© 1990 Optical Society of America