Research On Low-light Image Enhancement And Denoising Algorithm Based On Unsupervised Learning Multi-stream Feature Modeling

Low-light image refer to problems such as low brightness,unclear details,color distortion,and increased noise in images caused by insufficient lighting or dark environments.These issues lead to poor visualization of low-light image,and can also have a significant impact on some later visual tasks such as object detection,face recognition,medical image diagnosis,etc.Through low-light image enhancement algorithms,the quality and visual effects of images can be improved,making them clearer and brighter.These algorithms can be achieved by improving image contrast,reducing noise,and enhancing details.Low-light image enhancement has a wide range of applications in many fields,such as computer vision,medical imaging,security monitoring,drone imaging,etc.Therefore,the research on low-light image enhancement algorithms still has important significance.The main research work of this article is to address the problems of existing low-light image enhancement algorithms.1.A novel low-light image enhancement and denoising algorithm based on unsupervised learning multi stream feature modeling is designed.The input low-light image is enhanced through multiple branches,so as to extract its global and local features,and fully mine the spatial information and apparent features of the image,so that the enhanced image not only has rich detail information,but also can realize denoising and artifact removal.2.In the network generation stage,the Swin Transformer Block(STB)is innovatively used as the global feature extractor of low-light image.Its shift window mechanism can model the longdistance feature dependent of the input image with less network parameters,extract the color,texture and shape features,and suppress noise and artifacts.In the local feature modeling,this paper also adds a multi-scale image and feature fusion branch,which allows the exchange of information from different scales in U-net,so as to control the exposure of local areas on different scale images.3.In the discrimination network,the deep and shallow feature aggregation module is added to enhance the discrimination ability of the network.It uses the improved adaptive feature fusion mechanism to suppress the inconsistency by learning the contradictory information of the spatial filter,and realize the mutual guidance between the shallow representation information and the deep semantic information,so as to control the generation network make the generated image more natural,and improve the visual experience of human eyes.By introducing the above three innovative works,the low-light image enhancement and denoising algorithm proposed in this paper based on unsupervised learning multi stream feature modeling has achieved significant improvement in the low-light image enhancement performance.In testing experiments on multiple common datasets,the proposed methods have achieved better performance compared to some existing advanced low-light enhancement algorithms.