Research On Low-light Image Enhancement Algorithm By Integrating Global And Local Informatio

The low-light image enhancement task aims to enhance images in low-light environments,improve the brightness and contrast of images,and is the fundamental task of many computer vision tasks and has a wide range of applications.With the rapid development of deep learning in recent years,the model based on deep learning has been qualitatively improved in terms of the ability to extract information features from data.In this paper,using deep learning technology,aiming at the problems of insufficient image brightness and contrast,artifacts and blurring in some existing methods,a low-light image enhancement algorithm that integrates global and local information is proposed,and the global and local information of the image is realized to use of features efficiently.The main research contents are as follows:Most of the current low-light image enhancement work focuse on the improvement of image brightness,and less attention is paid to noise effects,which will affect the enhancement effect of low-light images.In response to this problem,this paper proposes a low-light image enhancement network model based on a contextual transformer model.The network mainly has three major modules.First,it uses the dynamic convolutional network to perform a shallow feature extraction on the input low-light image;then,it uses the contextual transformer module to perform global associated deep feature extraction,and uses the pyramid denoising module to remove the redundant information of the low-level features in the image.;finally,the enhanced image is obtained through the convolution module of the hourglass structure.This work is compared with the work in recent years on multiple datasets.The results prove that this method can effectively enhance low-light images,while preserving clear details and textures,removing noise,and achieving good performance in enhancing brightness and contrast.Most current low-light image enhancement methods focus on enhancing image brightness,while less attention is paid to handling noise,which can negatively impact the enhancement results.To address this issue,this paper proposes a low-light image enhancement network model based on a contextual transformer.The network consists of three parts: first,a dynamic convolutional network is used to perform shallow feature extraction on the input image;second,a contextual transformer module is used to perform deep feature extraction with global correlation,and a pyramid denoising module is used to remove redundant information in the image;finally,an enhanced image is obtained through a bottleneck structure convolutional neural network.The proposed method is compared with previous methods on multiple datasets,and the results demonstrate that this method effectively enhances low-light images while preserving clear details and textures,removing noise,and achieving good performance in enhancing brightness and contrast.Based on the first work,this paper addresses the insufficient processing of local information in the first work by designing a dual-branch feature enhancement network based on local-global information,which further fully utilizes the local and global feature information of the image.Specifically,a U-shaped network is designed in the local branch to perform multiscale feature extraction and fusion of the image’s local feature information using encoding and decoding modules.In the global branch,the global feature extraction ability and robustness of the Transformer module are utilized,and the model’s computational speed is increased while enhancing the model’s expressive power through a multi-head self-attention module based on the height-width direction.Finally,through the feature fusion module of the two branches,a channel attention mechanism is added to adaptively select information of different spatial scales,balance the feature weights of the two branches,and ultimately fuse and output the enhanced image.Compared with the first work,various evaluation indicators have been improved,and the obtained enhanced images can achieve better results,which proves that the method of this paper has certain advantages in the field of low-light image enhancement.