基于CNN-transformer轻量级网络的低光图像增强方法  

Low-light image enhancement method based on CNN-transformer lightweight network

作　　者：贺晨亚 崔学英[1] HE Chenya;CUI Xueying(School of Applied Science,Taiyuan University of Science and Technology,Taiyuan 030024,P.R.China)

机构地区：[1]太原科技大学应用科学学院,太原030024

出　　处：《灯与照明》2025年第1期118-120,共3页Light & Lighting

基　　金：山西省大学生创新创业训练项目(20230710);校级大学生创新创业训练项目(DCX2024185)。

摘　　要：由于设备、光照、拍摄条件等因素影响,导致拍摄的图片曝光不足出现暗区、褪色区域,影响图像中物体的识别。本文提出了一种轻量级的基于CNN-transformer的低曝光图像增强方法,用于图像的实时增强。该方法借助过曝光图像提供的信息,利用卷积模块和Transformer模块分别提取低曝光图像与过曝光图像的局部和全局信息,并将提取的局部信息与全局信息融合,以估计像素级的高阶曲线的动态范围来调整给定的低曝光图像,得到增强后的图像,实验结果说明了提出的方法的优越性。该技术可嵌入到手机和数码相机等设备中用于照片曝光强度的校正,满足实时性需求。Due to the influence of equipment,lighting,shooting conditions and other factors,the dark area and faded area appear in the underexposed picture,which affects the recognition of objects in the image.In this paper,a lightweight CNN-transformer-based low-exposure image enhancement method is proposed for real-time image enhancement.With the help of the information provided by the overexposed image,the convolution module and the Transformer module are used to extract the local and global information of the low-exposure image and the overexposed image respectively,and the extracted local information is fused with the global information to estimate the dynamic range of the pixel-level high-order curve to adjust the given low-exposure image to obtain the enhanced image,and the experimental results show the superiority of the proposed method.This technology can be embedded in devices such as mobile phones and digital cameras to correct the exposure intensity of photos to meet real-time requirements.

关 键 词：TRANSFORMER 图像增强 轻量级 双分支 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]