Research On Image Enhancement Algorithm Based On Convolutional Neural Network In Scotopic Vision Environment

There are two types of receptor cells in the human eye,cone cells and rod cells.These two types of cells divide vision into photopic vision,intermediate vision,and scotopic vision based on the brightness of the surrounding environment.Among them,the image collected in the scotopic vision environment will be insufficient due to the surrounding environment,resulting in the reflected light received by the device during the image acquisition process is weak,affecting the quality of the image imaging,so that the overall image appears dark,resulting in The human eye cannot directly obtain useful information in the image,which also affects the further processing of the image.Due to the problems of low brightness,low contrast,and unclear content in images in scotopic vision environments,convolutional neural networks have great advantages in image generation.Therefore,this thesis proposes an image enhancement algorithm based on convolutional neural networks in Scotopic vision environment.The main work of the thesis includes:1.Aiming at the problem of low image brightness in Scotopic vision environment.In this thesis,combined with the principle of Retinex algorithm,the collected image in Scotopic vision environment is decomposed to obtain illumination image and reflection image,the decomposed illumination image is reversed,and then the reversed image is connected to the image in Scotopic vision environment to input into the improved generation network,so that improve the overall brightness of the image;2.Aiming at the problem of image detail loss and pixel deviation in the process of image enhancement.This thesis combines the structural characteristics of U-Net network to improve the generated network.Combined with the characteristics of the human eye's visual perception process,a loss function based on the similarity of the image structure is added to make the image better meet the requirements of human visual;3.Aiming at the problem of image noise introduced in image acquisition and partial color deviation of enhanced image.In this thesis,Dn CNN denoising network is used to denoise the enhanced image and the reflected image of the image in the Scotopic vision environment,and then two denoised images are superimposed together to achieve the optimization of image denoising and color correction.Finally,in this thesis,the improved method is compared with the other four image enhancement methods,and makes subjective evaluation and objective evaluation respectively.The experimental results show that the brightness and details of the image obtained by the method proposed in this thesis are improved,and the overall visual effect of the image is also improved.