Abstract

The behavior of an optical processor architecture having three input images and one inhibited output image is described. It is applied as a major building block of an adaptive, self-organizing two-slab pattern classifier from the Grossberg neural models. Laboratory experiments are presented which indicate that the pattern enhancement, normalization, and inhibitory convolution effects required by the neural model may be achieved with a liquid crystal light valve such as the one produced by Hughes. This device is shown to give an intensity response similar to the fundamental nonlinear neural response required by the Grossberg models. A hybrid electrooptic design using current technology is proposed for the implementation of the complete adaptive pattern classifier.

© 1986 Optical Society of America