| With the rapid development of modern technology,the application of images in fields such as production and life,scientific research,etc.is also becoming increasingly widespread.However,in daily life,due to low light environments and technological limitations,captured images often face issues such as low brightness,noise interference,and lack of detail.Not only does it affect human vision,but it also causes difficulties in image processing in many fields,such as object detection,industrial production,remote sensing,and monitoring.Low light image enhancement aims to restore low light images to normal light images,enhance image brightness,reduce image noise,improve image quality,and provide high-quality image data for subsequent work.In recent years,low light image enhancement technology has made great progress,but there are still problems such as low contrast,over adjustment,and noise amplification after using existing algorithms to repair images.Therefore,low light image enhancement has become an urgent problem to be solved.This article proposes two effective methods to improve the quality of low light image,as follows:(1)An unsupervised low light image iterative enhancement network model is designed to address the issues of missing detail information and local information loss in feature extraction during low light image enhancement.Based on the Swin-Unet framework and image specific curve function iterative enhancement idea,the Swin Transformer Block is combined with the U-net network framework for feature extraction during image enhancement.Different scale features are fused to obtain rich global and local information,thereby solving the problem of missing detail information and color deviation during low light image enhancement,and effectively recovering image texture details.The experimental results show that the network structure does not require paired training data during the training process,and the model runs fast with low resource consumption.While effective improvement in luminance of low-light images,local texture information can be preserved.(2)In response to the problem of poor image enhancement performance under complex lighting conditions such as extreme darkness during low light image quality enhancement,a semi instance normalization module is introduced in the Swin Transformer Block of the feature extraction network Swin-Unet structure to restore shallow features of the image and improve the quality of the Real image.The experimental results indicate that this method solves the problem of insufficient texture details that may occur during low light image enhancement,and verifies the image enhancement effect under extreme dark conditions through comparative experiments.(3)Using the low light image enhancement method mentioned above,based on the functional requirements and feasibility analysis of the low light image enhancement application system,corresponding low light image enhancement application software has been designed and developed.By embedding a low light image enhancement algorithm based on the research content of this article into the system,the enhancement processing task of low light images and the display of image quality evaluation results can be achieved,meeting the practical application requirements of the unsupervised low light image enhancement method based on Zero-Unet.In summary,the zero reference low light image enhancement network based on Swin-Unet and Swin-HIN-Unet provides a new solution to the problem of low light image enhancement,and the implementation system designed and developed on this basis provides technical support for its practical implementation. |
