Research On Single Image Enhancement Algorithm Under Low-light Condition

In recent years,with the development of hardware such as smart phones,the number of personal photos has been increasing substantially.However,there are a large number of low-light images in a large number of personal photos,and the visual quality of these low-light images is not always guaranteed.In addition,computer vision and multimedia computing algorithms require highly visible images,but images taken in low-light conditions tend to be less visible.In order to solve the above problems,various single low-light image enhancement techniques have been proposed.But with the increasing visibility,existing technologies inevitably introduce various unnatural effects,such as excessively exaggerated details and color distortion.Based on the fact that in most cases people only have a single low-image available for enhancement,this thesis will start the study of single image enhancement algorithm under low-light conditions on the foundation of the existing low-light image enhancement technology.In order to solve the problem that the low-light region rarely occupies the entire area of the image in the previous study,and the problem of unnatural effect is introduced without how to adjust the enhancement intensity,this paper explores the two aspects of clarifying which areas need to remain unchanged and which areas need to enhance,and generating multiple homologous signals from a single input image to simulate multiple sources with different exposures.The main research contents of the thesis are as follows:(1)An image contrast enhancement model based on the simplified Retinex model with robustness under different illumination conditions is proposed.In order to seamlessly fuse the intermediate enhanced image with the original image as the final enhancement result,this work constructs a pixel-level weight matrix estimation and the work is based on the Retinex model,which automatically enhances dark areas and preserves bright areas.Multiple sets of qualitative and quantitative experiments verify the effectiveness of the proposed method,and show that this method can enhance image contrast while ensuring the naturalness of the image under different illumination conditions.(2)Next,a low-light image enhancement method based on multi-source fusion is proposed.Firstly,this work uses the simplified Retinex model to generate multiple enhancement results.Secondly,on the basis of these enhancement sources,the image decomposition and fusion model of the structural patch is adopted.After comparing with qualitative and quantitative experiments with several related methods,it is verified that this work can not only reduce the artifact effect and enhance the color performance of the enhanced results,but also generate enhanced results that can match with the image aesthetic evaluation better.