Research On Algorithm Of Low-Light Image Enhancement Based On Retinex

In low-light conditions such as nighttime,cloudy days and backlit shooting by the shooter,the captured images are prone to low brightness,significant noise and blurred details.Low-light image enhancement is a technique that can effectively improve the quality of lowlight images,give the human eyes a rich visual effect,and better pre-processing analysis for the digital image field.With in-depth research in the field of low-light image enhancement,many algorithms have emerged,but the existing methods are still prone to visual discomfort and difficult to provide high-quality data for image analysis.Therefore,there is a need to further improve the enhancement techniques for low-light images.This thesis aims to study low-light image enhancement techniques for the degradation of low-light images.On the one hand,it focuses on improving the brightness of the image to increase the visibility of the dark region;on the other hand,it pays attention to preventing over-enhancement of the image,maintaining the color naturalness of the bright region of the original image,overcoming the halo artifacts and suppressing the noise to meet the visual requirements of human eyes.This thesis has the following main work.(1)A Retinex variational model based on exponential form is proposed.The model is constructed by combining reflectance,illumination and bright channel a priori information,where the exponential form of the total variance and the exponential form of the local variation deviation are designed to better estimate the illumination and reflection components,and the model is solved by alternating iterative minimization method.Experimental results prove that the algorithm is more effective in improving the brightness,contrast,and detail information of images.The results are better than existing methods in terms of subjective and objective analysis.(2)A network for Retinex low-light image enhancement based on an attention mechanism is proposed.The Retinex decomposition model is designed to obtain the illumination and reflectance maps.To better smooth the illumination,a new attention mechanism module is introduced to obtain the adjusted illumination map,and finally the image is reconstructed by combining the reflectance map.The experimental results prove that the algorithm has appropriate contrast and overall conforms to the human eyes visual system,and local areas also avoid problems such as halo artifacts and unnatural edge shadows,which also show the advantages of the method in objective image quality assessment.