Biomimetic Computational Model Based On Visual Pathways

Visual perception is an efficient way for humans to perceive and understand the surrounding world.Humans receive outside optical signal through their eyes,undergo complex information processing and decoding processes,and ultimately form a perception and understanding of the surrounding environment.Compared with machine vision,the human visual system has many advantages,especially in adapting and reacting quickly in complex environments,with high robustness and adaptability.The visual system is crucial for many tasks in human daily life and work,such as recognizing objects,distinguishing colors and shapes,and sensing motion and depth.Therefore,studying the intrinsic working mechanisms of the biological visual system can not only help improve the ability of machine vision to process images,but also better understand the nature of human cognition.This article focuses on the perspective of bionic computing models,simulating the collaborative processing mechanism of multiple pathways in biological vision.By combining the contour detection,color constancy,and image dehazing applications in computer vision,feasible biological theories and effective implementation methods are provided for bionic computing models.The main research contents of this article are as follows:(1)A contour detection model based on Magnocellular/Parvocellular(M/P)dual path parallel mechanism is proposed.Firstly,the light sensitivity characteristics of retina are simulated,and a brightness gain model is constructed.Secondly,the processing process of the magnocellular pathway on brightness information is simulated,and a contrast extraction model of brightness and contrast fusion with fixed gaze micro-movement characteristics is constructed to suppress texture information.At the same time,the processing process of the parvocellular pathway on color information is simulated,and a color contour information extraction model is constructed using the center offset double-opponency receptive field model.Finally,the information fusion mechanism of the V1 area is simulated to achieve rapid contour information extraction.In the BSDS500 and BIPED datasets,the best single-image average P-index of this method is 0.42 and 0.50,respectively.This indicates that this method can effectively extract contour information and suppress texture noise.(2)A color constancy model based on mechanism of the center offset receptive field surrounding mechanism is proposed.Firstly,a center offset opponency receptive structure is constructed,and the size is adaptively adjusted using the contrast information extracted by the retina.Secondly,the color antagonism mechanism of the primary visual cortex is simulated,and the scene light source is initially estimated by combining the surrounding mechanism.Finally,simulate the integration of color through the ventral pathway and the color constancy of the advanced visual cortex to achieve color correction for color biased images.The median angular error on the SFU lab,SFU HDR,and GehlerShi dataset is2.4,2.9 and 2.7.Compared with other models,our model achieved good results and can effectively eliminate chromatic aberration and restore the true light source of the scene.(3)An image dehazing model based on the ON/OFF cell of primary visual pathway is proposed.Firstly,the horizontal cells and ON/OFF bipolar cells in the retina are simulated to remove atmospheric astigmatism while adjusting the contrast of foggy images.Secondly,the activity of the ON/OFF magnocellular and parvocellular is simulated to enhance the dynamic range of the defogging image.Finally,the foggy image is restored in the primary visual cortex of V1,while enhancing image details.In the NYU-Depth-V2 synthetic foggy image dataset,the peak signal-to-noise ratio(PSNR)and structural similarity index(SSIM)are 17.300 dB and 0.842,respectively.In the natural image dataset,the average gradient and information entropy are 44.231 and 7.354,respectively.Compared with other models,this model achieved good results and can effectively dehaze the image while improving image distortion issues.