Pattern Identification By Spatial Filtering In A Neuron Network Model

Abstract

Coherent optical signal processing techniques can be used to perform pattern identification test procedures. It is shown that there are four properties of the optical signal processing techniques that are necessary and sufficient for pattern identification. The purpose of this paper is to present a candidate neural pattern identification procedure and a realistic neuron network model able to execute the pattern identification procedure.

The identification procedure and neuron network model are obtained from an analogy between coherent optical systems and neuron networks. The four neural analogs of these optical properties are identified and shown to be reasonable neural properties. The four analogs between optical and neural systems compare and identify: (1) the neuron's pulse rate with the magnitude of a light wave, (2) the ‘phase difference’ between any two synapses with the phase difference between any two points in the coherent light beam, (3) the spatial dispersion of the neuron's pulse train with the optical dispersion of light from a point source and (4) the variation in evoked neural response to identical stimuli with the filter properties of the optical system.

While it has not been possible to show the neural identification procedure is implemented by filtering the visual input pattern in the optical sense, the principal objective has been to show that a neuron network model is able to implement this identification procedure. Experimental evidence tends to support this neuron network model as an element of the visual processor in cats.