Research On Computer Image Recognition Based On Perception Theory

For computer vision problems,there were two relative concepts as early as the forties or fifties of the 20 th century,which were “from local to global” and “from global to local”.One of them was the Marr-based local characteristics determine the overall view,and another think that visual perceptual process is the global priority over the local based on Gestalt theory.The view of that local decisions the global has been dominant for half a century,until Chen Lin put forward the theory of topological perception in the “Visual Cognition” in 2003.He specified defined that “Topological properties is prior to local features”.This article make research on the relationship between human brain and computer image recognition that based on the perception theory.Two aspects will to be analyzed.One is the visual perception theory,and the other is cognitive biology.From the perspective of visual perception theory,this article mainly discusses two milestone theories there are the theory of Gestalt and Chen Lin's theory of topological perception.From the perspective of cognitive biology,this article mainly describes the deep learning network that's based on the simulations of biological neural network structures.The human brain directly affects the visual cognitive process by studying the relationship between the human brain's structure and the cognitive process.Firstly,this article studies development process of human visual cognitive ability.Through the research,this article think initial identification of the initial stage of the human visual system is the identification of the color and region,and color is the earliest feature of human visual system.At the same time,the attention of the global is also the characteristics of the human visual system.Second,the Gestalt theory and topology perception theory are integrated on classic problem of "split object and background".At the same time the "Significance-Choice-Gestalt" idea of dealing with the visual problems were perfected.Based on the human brain topology and visual perception process,this article presents the algorithm of subject's contour extraction.The theoretical knowledge,implementation basis of the algorithm and the whole algorithm model are described in detail.Then comparision in the extraction of the main area about the main contour extraction algorithm and GrabCut algorithm,and the practical application of the subject's contour extraction algorithm in cats and dogs photo classification,which prove the usability of the algorithm.It is proved that through tracing and testing cats and dogs photos using the AlexNet-based depth learning neural network,learning and training the datas generated by the processing of this algorithm is much better than directly evaluating the original images.On the one hand,for the processing of small data sets this algorithm has the advantage a deep learning network that does not.On the other hand,the accuracy of identification classification is improved.