Sensory neural networks and their adaptations

The signal processing principles utilized in sensory neural systems such as the retina and optic lobes of insects are studied with a view to deriving principles for design of analog integrated circuit peripheral (early) processing systems. The basic peripheral processing principles in biological visual systems are gain adaptation and lateral inhibition. The utility, both in biological and artificial neural networks, of nonlinear lateral inhibition is examined. Because lateral inhibition involves only near neighbor interactions between cells, optoelectronic implementations in the form of monolithically integrated photodetector arrays are particularly attractive. These have wide practical applicability for machine vision, optical scanning, and image enhancement purposes. Such implementations are completely analog and can be designed to require only one transistor per unilateral interconnection.<>