On a problem of relations between pattern recognition, pattern formation and moving multipoint objects perception

A novel approach to the visual system pattern recognition synthesis is proposed. The nonlinear net with continuous activating parameters representing positions of the perceived set of dots in the Euclidean space is developed based on the principle of geometrical invariant pattern formation. This net memorizes patterns by their relative location towards the system centroid point, which allows the net to distinguish objects that may be naturally realized from those impossible states. As an effect of the net the properties of distinctly moving dots gathering and resulting structure recognition in the independent coordinate space appear. At the same time the proposed model does not recognize individually moving dots as the structure if the motion is an accelerated process