The Research Of Visual Feature Integrate Model Based On Visual Mechanism And Its Application

With the demand of image-related control technology which growth rapidly in the fields of social, economic, national security and so on, it becomes a very pressing problem to develop advantages among information science, life science and mathematical science, and find the deep information processing mechanism in the visual system to improve the image information understanding ability and treatment efficiency of computer.The efficiency of animal visual system mainly reflects in the way of information processing of visual cortex cells, and not in the rapidity of signal conduction. In this paper, we take the efficient information processing mechanism that exist in early stage of visual system as breakthrough point, explored the information integrate mechanisms of primary visual cortex according to information processing characteristics among hierarchical, parallel and interact between cortex. We proposed a computable model and designed key algorithm of the model. According to some specific applications, we tested the feasibility and effectiveness of the model and algorithm. The main research results as follows:(1) Analyzed the transfer process and key information processing mechanisms of visual information; start from statistical characteristics of natural images to establish the simulate model of visual information processing; research the classical linear mode and its main optimize criterion and optimize algorithm from the spatial association, high-order statistical characteristics and Space-time Characteristics of natural images.(2) To simulate the feature extract and integrate mechanism of visual system, we proposed a computable model of visual feature integration. The model is consisting of three parts that are retinal pre-process, receptive fields'initial integration and synchronize integration. The image feature was pre-processed in retina firstly to remove second-order redundancy, and then extracted topological features of images by receptive fields'initial integration. Finally, according to the test image, we do synchronize integration to image features and select the best neural response. According to computable model, we designed the corresponding key algorithm.(3) For the test of the model and algorithm, we do three applications that are train crackles detection, traffic vehicle detection and concealment target search. Experimental results show that the algorithm is better than traditional methods at the rate of crack detection, vehicle detection accuracy and target acquisition accuracy.(4) According to key algorithm, we design and implement a prototype system of visual feature integrate which based on the MATLAB 7.0. The system can complete the identification and detection of the above three type's application.