基于生成对抗网络的弱光图像增强方法  

Low-light image enhancement method based on generative adversarial network

作　　者：武霁 丁冰 丁洁[2] WU Ji;DING Bing;DING Jie(College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,China;School of Integrated Circuits and Electronics,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]太原理工大学电气与动力工程学院,山西太原030024 [2]北京理工大学集成电路与电子学院,北京100081

出　　处：《现代电子技术》2024年第15期53-59,共7页Modern Electronics Technique

基　　金：国家重点研发计划(2022YFB3204600);北京理工大学青年教师学术启动计划。

摘　　要：现有的弱光图像增强方法大多存在色彩失真,去噪效果不佳,严重依赖成对数据集进行训练等问题。针对上述问题,提出一种基于生成对抗网络的弱光图像增强方法。该模型分为生成器和判别器两部分,生成器部分使用添加EMA注意力的改进UNet网络进行图像增强,判别器部分采用包括颜色判别器、灰度判别器和多尺度判别器的多分支判别器进行融合判别图像的真实性。实验结果表明,文中方法在公开数据集上取得了优异的效果,在PSNR、SSIM、NIQE、BRISQUE等多项评价指标上有了显著提升,进一步证明了所提方法的有效性和鲁棒性。Most of the existing low-light image enhancement methods suffer from issues such as color distortion,poor denoising performance,and heavy reliance on paired training datasets.To overcome these challenges,a low-light image enhancement method based on generative adversarial network is proposed.The proposed model comprises two parts,named a generator and a discriminator.In the former part,the improved UNet network with integrated EMA(efficient multi-scale attention)is used to enhance the image,while in the later part,the multi-branch discriminator including color discriminator,grayscale discriminator and multi-scale discriminator is used to fuse and judge the authenticity of the image.The experimental results show that the proposed method achieves excellent results on public datasets,and shows significant improvements in evaluation indexes such as PSNR(peak signal-to-noise ratio),SSIM(structural similarity index measure),NIQE(natural image quality evaluator)and BRISQUE(blind/referenceless image spatial quality evaluator),which further proves the effectiveness and robustness of the proposed method.

关 键 词：弱光图像 无监督学习 生成器 判别器 注意力机制 图像增强 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]