Enhancement And Restoration Of Low Quality Images

With the development of technology and the popularity of the Internet,people tend to share various pictures through the network.In order to reduce the storage and increase the transfer rate of images,these pictures are often compressed.And then the compressed image would produce a certain degree of compression artifacts.When a compressed object is a low-light image,its compression artifacts are generally hidden in the dark areas of the image.After the brightness is enhanced,the original dark area will produce more obvious compression artifacts,which affects the visual effect of the image.Therefore,in order to improve the quality of the low-light compressed image,and provide a better visual experience to the users,it is necessary to reduce the compression artifact of the image while enhancing the brightness of the low-light compressed image.Based on this,this paper investigates the related algorithms of low-light image enhancement and reduction of compression artifacts for the enhancement of low-light compressed images,and finds that how the order of these algorithms(first brightness enhancement then reduce artifacts or reduce artifacts first)are used,the effect of processing is flawed.When the low-light compression image is reduced first and then the brightness is enhanced,the result is excessive smoothing;on the contrary,the result of the artifact suppression effect is not obvious,because the image compression is pseudo during the brightness enhancement process.The shadow has also been enhanced,increasing the difficulty of artifact suppression.Taking into account the above problems,this paper adopts the idea of divide and conquer,and proposes a method of image enhancement based on image decomposition for low-light compression,which achieves low-light enhancement while suppressing compression artifacts.The main work of this paper is:(1)This paper uses a guided filter to extract the illumination layer of the image and based on the Retinex theory.We firstly calculate the reflective layer of the image.Then,the gamma transform is used to enhance the image brightness on the illumination layer,the total variation regularization method is used on the reflection layer to reduce the compression artifact,and finally,the processed illumination layer and the reflection layer are recombined.The research objectives of this paper have been initially achieved.(2)Based on the framework proposed in this paper,the related processing methods of(1)are optimized.Use a more flexible and applicable decomposition model,which eombine with the idea of image fusion to achieve image brightness enhaneement.In addition,this paper attempts to use the deep learning method to reduce the compression artifacts of the image and compare it with the method of total variation regularization.Finally,the reflectance layer after the artifact reduction is combined with the illumination layer after the brightness enhancement to obtain the final output result.And through experiments,the effectiveness of the method is proved.