Research On Multilayer Visual Network Model For Image Processing Applications

In the field of image information analysis and understanding,biological vision has extraordinary capabilities far beyond machine vision.Through the research and simulation of the operating mechanism of its nervous system,on the one hand it will help to better understand the process of visual information transmission and processing,on the other hand,it will provide the necessary theoretical foundation for the construction of visual network model.This will bring a new research idea for the application of image processing algorithms.This paper considered the characteristics of visual information transmission,constructed the corresponding multi-layer visual network model based on the visual response characteristics of multi-layer mechanism in the visual pathway,and proposed the visual fusion model of multi-visual pathways.This paper studied the rules and mechanisms of visual information processing,introduced a retinal neuronal network and a high-level neural feedback unit of the visual cortex for multi-layer visual model in the visual pathway.In this paper,contour detection was taken as an example,and the multi-layer visual network model was applied to image processing.The main research work and achievements of this paper are as follows:(1)This paper improved the classical receptive field response of ganglion cells in the visual pathway,proposed the negative effect of DOG on the antagonistic mechanism of the ganglion cells.In this paper,an orientation perception model of simple cell that can deal with this negative effect was designed,and a multi-path parallel mechanism of visual information processing was introduced,We divided the visual path into the main path and the secondary path,and put forward the method of using the difference of visual information in different path to suppress the texture and enhance the contour.The multi-path visual fusion model of complex cells in V1 area was designed.The experimental results show that the average P value of the proposed method is 0.45,which has achieved good results.The results were compared with those of DG,CORF and SSC.Compared with the contrast method,the proposed method eliminates most of the texture disturbances in the target contour detection and has better contour detection accuracy,which verifies the effectiveness and feasibility of the new method.(2)A new model of simple cell receptive field with double oval structure was proposed,which combined the characteristics of the receptive field of the multiscale ganglion cells,the visual attention and the feedback regulation of the high-level neural center.In this paper,a texture suppressor which can reflect the discrete degree of local images was introduced,which regulated the visual input of simple cells in primary visual cortex,and completed the task of contourextraction.In the comprehensive contour detection test of Ru G40 database and BSDS300 database,the average P-value index of the method proposed in this paper is 0.63.This method used the computational model with the physiological characteristics of the bottom of the biological vision to obtain the contour detail information and suppressed the texture by using the high-level neural center feedback regulation mechanism of the visual cortex,thereby retaining more contours and improving the overall detection performance.(3)In this paper,aiming at the complexity of texture noise of medical image,an improved visual computing model considering non-classical receptive field adjustment and feedback adjustment mechanism was proposed,which enhanced the texture suppression and noise adaptability.This paper selected medical erythrocyte images,E.coli images and lung CT images as experimental images,and the traditional edge detection method was used as a contrast.The experimental results show that the proposed method has some adaptability to the noise in medical images and has certain advantages in the detection of contour details.