For The Neural Computation Of Visual Information Processing

Visual computing is a comprehensive and challenging subject. So far, plentiful fruits have been got in that research. However, people also find that the difficulty is much greater than what's expected in the process of exploring visual perception more. That difficulty roots in the localization of the understanding of visual mechanism and technical restrictions to visual computing. So, people must improve their understanding of the vision and renovate their computing means in order to solve the problems.Advancements from visual neural science make the principle of the vision understood more deeply and also promote researches of visual computing. Neural network is a bottom-to-up method of simulating intelligence. Because of its innate advantage, it is able to import physiological characteristics into visual computing better and more easily, as favors solutions to a lot of problems.This paper does researches on some typical problems of visual computing with neural network, whose work includes,(1) Analyses to physiological characteristics of visual system. It introduces some of them that is related to the paper and expatiates the strategy of the simulation.(2) Edge detection based on neural dynamic system. It analyzes all kinds of dynamic characteristics of continuous Hopfield network. The methods of edge detection on that simplify the linking of the network and accelerate the process.(3) Structural analyses to the image. It defines basic structure of the image and presents means of detecting them. The method of edge optimization based on structural idea makes the contradiction between eliminating noises and reserving details solved better.(4) Region division based on neural dynamic system. It presents means of serial division and integral division by continuous Hopfieldnetwork. The former has the advantage of topology invariability and the latter can accomplish the division well when there are gaps in the edge.(5) Fuzzy neural reasoning. It expatiates neural expressions of association relation and some logic relations and establishes corresponding neural reasoning machines. With helps of universal logics and evolution computing, it constructs the model of the combination of multiple neural reasoning machines, as solves the problem of the cooperation of reasoning systems from different domains.Gupta's tri-graded model is the basis of the paper, which is the abstract of human visual system.How to express and find structural information and how to organize and utilize transcendent knowledge have puzzled researchers of visual computing all along. Such questions are discussed and some new methods are presented in this paper. It is a good start, however, deeper and more extensive researches will be needed in the future.