Abstract

Human vision employs principles that differ significantly from existing technological pattern recognition systems. Yet, based on existing models, the biological process may be broken down into a series of steps that map well onto parallel optoelectronic implementations. In particular, the Grossberg-Mingolla boundary contour systems (BCS) model of vision will be examined here. The principal feature of the BCS system is its ability to complete boundaries between image regions that differ in luminance or textural distribution. This model will be explained and then interpreted in terms of cascaded stages of correlation that may be implemented optically. Verification of this interpretation will be provided through simulations of its performance on complex scenes. Translation of the biological model to a correlator model raises a number of engineering issues. These will be addressed, and their resolution will be explained.

© 1992 Optical Society of America