Research On Low-light Image Enhancement Methods

The camera can't always capture high-visibility images,due to lacking of light,uneven illumination,weather changes,relative motion,overexposure and other factors,which leads great challenges on subsequent machine vision applications.Therefore,it is necessary and imperative to post-process the image so that it still has high quality standards in a complicated environment.Low-light color images not only have the characteristics of low brightness,but also are accompanied by noise,color shift,etc.due to reasons such as equipment accuracy.In addition,it is easy to amplify noise,color shift,halo artifacts,and unnatural gradient order during the image enhancement process.Aiming at the problems of low-light image enhancement,the research work mainly includes the following three parts.First,we proposed a robust and comprehensive enhancement method based several points.First,the idea of bright channel is introduced to estimate the illumination map which is used to enhance the image with Retinex model,and the color constancy is keep as well.Second,multi-scale morphological closing operation is used to modify the erroneous estimating illumination.Finally,guided filter is used to correct the gradient edges to avoid halo artifacts.The proposed algorithm can enhance the visual characteristics of the darker part,keep the color constant,no halo artifacts,and no overexposure.The method is mainly used to enhance uneven illumination images in lowilluminance images,but its enhancement effect on other types are not stable.Second,we present a comprehensive image enhancement method for low-quality illumination environments which can enhance image,keep naturalness,compensate saturation as well as.First,dilation-limited is used to estimate the global illumination intensity.Second,for keeping the naturalness,the small patch closing operation is used to estimate the illumination edges first,the guided filter is redefined to fusion the illumination intensity and edges next.Furthermore,adaptive saturation fusion algorithm was proposed to recover the lost color information.The superior performance of this method has been demonstrated by comparison other state-of-the-art methods from subjective evaluation and objective evaluation.The method is improvement of the first part.It is more robust for all low-light image enhancement effects compared with the first part,but the method does not remove the noise in the image.Last,we proposed a low-quality image enhancement solution based on deep learning.First,we produced a data set for low-light image enhancement based on the fivek?data data-set;Second,modify the network on the basis of the U-Net network to make it suitable for the enhancement of low-quality images;This solution can effectively enlarge the features and textures hidden in the dark,and we added a noise removal module to the algorithm to eliminate the noise and reduce the influence of noise on the enlarged texture features.