Research On Visual Perception Mechanism And Its Application

The human visual system has extremely powerful information analysis and cognitive understanding,and is an important way for humans to perceive the world.In-depth study of the internal mechanism of the visual system and its modeling not only helps to understand the information processing and transmission methods of visual perception,but also contributes important theoretical foundations and new ideas for the realization of computing models based on visual perception mechanism.This paper first considers the close relationship between the orientation selectivity of the front and rear neurons,and proposes to extract the orientation information of the visual stimulus layer by layer from the depth selection model,and introduces the data-driven visual attention to measure the spatial sparsity of the visual information.The coding realizes the contour information enhancement;then the dynamic adaptation of the visual stimuli contrast changes by the primary visual cortex neurons is constructed,and the contrast adaptive orientation selection model is constructed to accurately capture the light and dark changes and orientation information of the visual information,and suppressing the texture noise by the non-classical receptive field modulated by the pre-stage suppression information;finally,based on the scale change of the parallel channel of the visual pathway,the color antagonism and the feedback control of the advanced cortex on the primary visual cortex are introduced to realize the image prominent contour detection and applied to the lane in the intelligent traffic scene.The line is automatically detected,and the results verify the feasibility of the method.The main research work and results of this paper are as follows:(1)Considering the orientation selectivity of the multi-level receptive field of the main visual pathway to the visual stimuli and the spatial position information sensitivity of the sub-visual pathway,a new contour detection method based on the synergy of the primary and secondary visual pathways is proposed.Aiming at the continuity of the contour line and the overall orientation,it is proposed to extract the orientation information in a multi-level manner towards the depth selection model;and use the neuron coded visual sparse information to simulate the feedforward visual attention process to achieve contour enhancement and noise filtering,and then complete the contour perception.Taking the natural scene image in the RuG40 data set as the detection target,the average P value of the detection result is 0.47,which has better detection performance than the comparison method.(2)To study the information processing mechanism of contrast adaptation of neurons and field suppression of non-classical receptiveness,a new contour detection method based on contrast adaptation and side suppression information is proposed.According to the dynamic adaptation and orientation selection characteristics of the visual cortex neurons,the difference of visual stimuli is detected and mapped into the response carrying the contour intensity information;the difference between the external geniculate field and the non-classical receptive field is compared.The primary visual cortex side inhibition intensity is globally regulated,protecting weak contours and suppressing texture noise.In the experiment against the RuG40 dataset,the average P-value index of this method is 0.48,which has played an excellent role in the information processing of the biological vision system.(3)A new contour detection method based on parallel channel grading processing of visual pathway is proposed.The large-scale information channel that quickly transmits low-frequency information is used to represent the overall contour,and the color-sensitive color information processing channel is introduced to respond to small-scale information,characterize the detail contour,and use the neural network coding to remove redundant information,and protect the contour through feedback control of the advanced cortex,then combined with large-scale and small-scale information to complete contour perception.The method is applied to the lane line automatic detection task,which effectively solves the problem of texture noise interference,and can better detect the lane line contour in the case of low illumination difference and low road surface cleanliness.