Research On Visual Information Coding Mechanism And Its Application

The extraordinary performance of the visual nervous system is closely related to the visual information coding mechanism,but people usually only focus on physiological and psychological based visual nerve function detection and construction,or just around the traditional information theory coding mechanism to start applied research.Although the above studies are also of great research significance,they lack the visual perspective of visual system characteristics and visual information coding mechanism,it is difficult to achieve the application of visual processing breakthrough.Therefore,this dissertation focuses on the visual information coding mechanism,constructs the visual path model,takes advantage of the visual system hierarchy and coding features,and attempts to apply it in the image processing represented by contour detection.In this paper,firstly,we use the receptive field of visual cortex neurons to dynamically adjust the response information of the anterior direction to construct the visual cortex double antagonistic receptive field model to select the direction of visual stimuli.The synaptic connection of neuronal population discharge coding enhances the contour information response;Secondly,in order to improve the rapidity of contour detection,a visual information processing method of multi-scale non-classical receptive field and visual cortex feedforward and integration is proposed,which can effectively highlight the contour details and restrain the background texture.Finally,the neuronal color constant feature is applied to image enhancement,and the directional feature of the image is coded hierarchically.The contour detection of foggy road image is used as a specific application,and the result is more feasible.The research and results of this paper mainly include three parts:(1)The information processing characteristics of visual color antagonism and dynamic coding of neuron population were studied,and a new method of contour detection based on neuronal color antagonism and dynamic coding was proposed.Firstly,a dynamic regulation method of single antagonistic receptive fields of ganglion cells and lateral geniculate bodies was proposed.The model of polar dendrites receiving polar distribution was established to realize the primary visual cortex response to orientation and spatial selection.Synaptic dynamic connection,fusion of single-cell pulse frequency response,to achieve the inhibition of texture information.BSDS500 database was used in the experiments,the mean value of measure P for 100 images was 0.58 ?0.04,relative to other contrast method,improves the accuracy of contour extraction.(2)A new contour detection method based on multi-scale non-classical receptive field feedforward and integration is proposed by considering the ability of non-classical receptive fields to suppress background texture at different scales and integrating the information of visual cortex feedforward.Aiming at the multi-contour natural images,a multi-scale non-classical receptive field modulation method is proposed,which can restrain tedious visual information and weaken the redundant texture.And improve the visual cortex integration of superior and subordinate information feed-forward mode of action to further complete the perception of the outline.Taking the natural background image as the experimental object,the value of measure P for 100 images was 0.61?0.05,compared with other methods can improve the accuracy of contour extraction(3)A new method of contour detection based on neuron color constant and hierarchical coding is proposed.From the color constant correction of color shift image and the rapid and accurate requirement of detection method,a model of dynamic adjustment of color shift image of neuron receptive field on visual pathway was established,the process of visual integration of visual cortex neurons was gradually simulated,and the information processing was selected.Target contour under abnormal environment.It is shown that the method can highlight the targets in the foggy road and improve the detection efficiency greatly.It is an important project for traffic monitoring system or driverless technology in the target recognition,scene understanding process in providing the theoretical basis.