Bioinspired Approaches to the Selection and Processing of Video Information

Abstract

Significant expansion of the range of applications of real-time video systems requires further improvement in their productivity, efficiency and intelligence. Therefore, researchers are increasingly turning to the human eye analyzer as a prototype to create more sophisticated systems of technical vision. The paper proposes a number of approaches and methods for the selection and processing of video information inspired by the human visual analyzer. In particular: the method of hierarchical selective perception of video information; dynamic models of processes for finding objects, tracking them, panning the scene and the mechanisms of attention and allocation of the essence, which, by managing the parameters of reading information from a video sensor, provide reading of a part of the image relevant to the task; information measure of the dynamic image (δ-entropy), which characterizes its spatial frequencies and is an effective information feature for the search and recognition of objects; methods of expanding the dynamic range of perception of brightness; the principles of circular organization neurons of the central fovea, which provide increased contrast and the allocation of informative features; circular organization of the retinal neurons with a sumation of signal sticks that contribute to increased sensitivity in conditions of insufficient lighting; specialization of neurons and organization multilayer neural network.