Neural-network model of artificial vision for pattern analysis

This paper proposes a model of artificial vision that takes into account the general principles for constructing and operating biological prototypes for solving problems of pattern analysis. The model is based on associative-processing methods, implemented by a two-layer optical neural network (NN) with a correlation matrix of connections. The requirements on the filtering function in the correlation layer that implements the selective-attention function are determined. To ensure that the selective-attention function can be adapted to an input image, the image layer is implemented as a layer for comparing an object image and the counter-expectation presented by the NN. (C) 1997 The Optical Society of America.