Design And Implementation Of Video Low-light Enhancement System Based On Transformer

In the practical use of mobile phone cameras,due to shooting environment or technical limitations,the collected videos often suffer from problems such as low brightness,uneven illumination,and color deviation,which are commonly referred to as low-light problems.Applying low-light enhancement algorithms can improve the brightness,signal-to-noise ratio,and contrast of low-light videos to enhance the user experience.Based on a comprehensive analysis of existing low-light enhancement technologies,this thesis proposes a Transformer-based low-light video enhancement algorithm and develops an iOS application with real-time video lowlight enhancement capabilities.The main contributions of this thesis are as follows:(1)A Transformer-based low-light video enhancement algorithm is proposed.Firstly,we design a video low-light enhancement network model,which includes two branches:a detail recovery network based on U-Net and a global brightness adjustment network based on Transformer.Secondly,this thesis proposes a lowlight dataset construction method based on inverse ISP(Image Signal Processor)and constructs an artificial synthetic paired dataset containing 20,000 pairs of images.Finally,a comprehensive comparative experiment is carried out,and the results show that the proposed algorithm has significant advantages in terms of PSNR,SSIM,model size,and computation time,and can effectively improve overall brightness while retaining and restoring image details.(2)An iOS application with real-time video low-light enhancement capabilities is designed and implemented.Firstly,a feasibility analysis,functional requirements analysis,and non-functional requirements analysis are conducted for video low-light enhancement requirements.Secondly,a concise and clear design style is adopted for the iOS application,and the Swift language is used for client development on the Xcode platform.The test results show that the application can enhance low-light videos in real time,offline,and stably,demonstrating effectiveness and ease of use.