基于渐进递归的生成对抗单幅图像去雨算法

doi:10.13229/j.cnki.jdxbgxb.20230756

吉林大学学报(工学版) ›› 2025, Vol. 55 ›› Issue (4): 1363-1373.doi: 10.13229/j.cnki.jdxbgxb.20230756

基于渐进递归的生成对抗单幅图像去雨算法

刘广文¹(),赵绮莹¹,王超²,高连宇¹,才华¹(),付强³

^1.长春理工大学电子信息工程学院，长春 130022
^2.长春理工大学空间光电技术国家与地方联合工程研究中心，长春 130022
^3.长春理工大学空间光电技术研究所，长春 130022

收稿日期:2023-07-18 出版日期:2025-04-01 发布日期:2025-06-19
通讯作者: 才华 E-mail:lgwen_2003@126.com;caihua@cust.edu.cn
作者简介:刘广文（1971-），男，副教授，博士.研究方向：信号与信息处理技术.E-mail： lgwen_2003@126.com
基金资助:
国家自然科学基金重大项目(61890963);吉林省科技发展计划项目(20210204099YY)

Progressive recursive generative adversarial network-based single-image rain removal algorithm

Guang-wen LIU¹(),Qi-ying ZHAO¹,Chao WANG²,Lian-yu Gao¹,Hua CAI¹(),Qiang FU³

^1.School of Electronic Information Engineer，Changchun University of Science and Technology，Changchun 130022，China
^2.National and Local Joint Engineering Research Center for Space Optoelectronics Technology，Changchun University of Science and Technology，Changchun 130022，China
^3.School of Opto-Electronic Engineer，Changchun University of Science and Technology，Changchun 130022，China

Received:2023-07-18 Online:2025-04-01 Published:2025-06-19
Contact: Hua CAI E-mail:lgwen_2003@126.com;caihua@cust.edu.cn

摘要/Abstract

摘要：

为了解决传统生成对抗网络（GAN）在去雨问题中因网络容量不均衡而表现不佳的问题，提出了一种基于渐进递归的生成对抗单幅图像去雨算法。该方法采用了渐进递归模块生成器和多尺度特征模块鉴别器，旨在提高生成器效率并增强鉴别能力。渐进递归模块通过融合多尺度特征并构建渐进递归结构，不仅减轻了网络参数的负担，也提高了生成器的效率。与此同时，多尺度特征模块有助于鉴别器从局部和全局两个层面提取特征，从而增强了其鉴别能力。实验结果表明：相较于现有算法，本文算法在Rain100L数据集上峰值信噪比（PSNR）和结构相似性（SSIM）指标提高了1.11%和1.16%；在Rain100H数据集上将指标分别提高了3.28%和1.01%；在真实数据集上，本文算法在去雨任务中表现出色，成功地保留了大部分细节特征。这些实验结果充分验证了本文算法的有效性和鲁棒性。

关键词: 计算机视觉, 图像去雨, 多尺度, 生成对抗网络, 渐进递归

Abstract:

To address the issue of traditional GAN networks underperforming in single-image rain removal due to imbalanced network capacity， this article introduces a progressive recursive generative adversarial algorithm for this task. This method employs a progressive recursive module generator and a multi-scale feature module discriminator， aiming to enhance the efficiency of the generator and bolster the discriminator's capability. The progressive recursive module， by merging multi-scale features and constructing a progressive recursive structure， not only reduces the burden of network parameters but also elevates the generator's efficiency. Concurrently， the multi-scale feature module aids the discriminator in extracting features at both local and global levels， thereby amplifying its discriminative power. Experimental results indicate that， compared to existing algorithms， our method achieves a peak signal-to-noise ratio （PSNR） and a structural similarity index measure （SSIM） were improved by 1.11% and 1.16% on the Rain100L dataset. On the Rain100H dataset， these metrics were improved by 3.28% and 1.01%， respectively. On real-world datasets， our algorithm excels in rain removal， successfully preserving the majority of detailed features. These experimental outcomes thoroughly verify the effectiveness and robustness of our proposed algorithm.

Key words: computer version, image rain removal, multi-scale, generative adversarial networks, progressive recursive

中图分类号:

TP391

刘广文,赵绮莹,王超,高连宇,才华,付强. 基于渐进递归的生成对抗单幅图像去雨算法[J]. 吉林大学学报(工学版), 2025, 55(4): 1363-1373.

Guang-wen LIU,Qi-ying ZHAO,Chao WANG,Lian-yu Gao,Hua CAI,Qiang FU. Progressive recursive generative adversarial network-based single-image rain removal algorithm[J]. Journal of Jilin University(Engineering and Technology Edition), 2025, 55(4): 1363-1373.

图/表 12

图1

图2

图3

图4

表1

图5

图6

表2

表3

表4

表5

损失函数参数设置对比"

参数	PSNR	SSIM
$α = 110$ ， $β = 1$	35.19	0.935
$α = 100$ ， $β = 10$	35.50	0.962
$α = 60$ ， $β = 50$	34.92	0.949
$α = 10$ ， $β = 100$	34.68	0.947

表5

表6

参考文献 24

[1]	陈舒曼, 陈玮, 尹钟. 单幅图像去雨算法研究现状及展望 [J]. 计算机应用研究, 2022, 39 (1): 9-17.
	Chen shu-man, Chen wei, Yin Zhong. Research status and prospect of single image ram removal algorithm[J]. Application Research of Computers, 2022, 39(1): 9-17.
[2]	Sun P, Zhang R, Jiang Y, et al. Sparse R-CNN: end-to-end object detection with learnable proposals[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), Nashville, USA, 2021: 14454-14463.
[3]	Carion N, Massa F, Synnaeve G, et al. End-to-end object detection with transformers[C] ∥Computer Vision-ECCV 2020: 16th European Conference, Glasgow, UK, 2020: 213-229.
[4]	Fu Z, Fu Z, Liu Q, et al. SparseTT: visual tracking with sparse transformers[J/OL].[2023-07-05]..
[5]	Song Z, Yu J, Chen Y P P, et al. Transformer tracking with cyclic shifting window attention[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, USA, 2022: 8791-8800.
[6]	Yan W, Tan R T, Yang W, et al. Self-aligned video deraining with transmission-depth consistency[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, USA, 2021: 11966-11976.
[7]	Yue Z, Xie J, Zhao Q, et al. Semi-supervised video deraining with dynamical rain generator[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, USA, 2021: 642-652.
[8]	Li Y, Tan R T, Guo X, et al. Rain streak removal using layer priors[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 2736-2744.
[9]	Kim J H, Sim J Y, Kim C S. Video deraining and desnowing using temporal correlation and low-rank matrix completion[J]. IEEE Transactions on Image Processing, 2015, 24(9): 2658-2670.
[10]	Jiang T X, Huang T Z, Zhao X L, et al. A novel tensor-based video rain streaks removal approach via utilizing discriminatively intrinsic priors[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,Honolulu, USA, 2017: 4057-4066.
[11]	Zhang H, Sindagi V, Patel V M. Image de-raining using a conditional generative adversarial network[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2019, 30(11): 3943-3956.
[12]	Qian R, Tan R T, Yang W, et al. Attentive generative adversarial network for raindrop removal from a single image[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, USA, 2018: 2482-2491.
[13]	Wei Y, Zhang Z, Wang Y, et al. Deraincyclegan: rain attentive cyclegan for single image deraining and rainmaking[J]. IEEE Transactions on Image Processing, 2021, 30: 4788-4801.
[14]	Metz L, Poole B, Pfau D, et al.Unrolled generative adversarial networks[J/OL].[2023-07-05]..
[15]	Fu X, Huang J, Zeng D, et al. Removing rain from single images via a deep detail network[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA, 2017: 3855-3863.
[16]	Yang W, Tan R T, Feng J, et al. Deep joint rain detection and removal from a single image[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, USA,2017: 1357-1366.
[17]	Zhang H, Patel V M. Density-aware single image de-raining using a multi-stream dense network[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,Salt Lake City, USA, 2018: 695-704.
[18]	Li X, Wu J, Lin Z, et al. Recurrent squeeze-and-excitation context aggregation net for single image deraining[C]∥Proceedings of the European Conference on Computer Vision (ECCV),Munich, Germany, 2018: 254-269.
[19]	Ren D, Zuo W, Hu Q, et al. Progressive image deraining networks: a better and simpler baseline[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, USA, 2019: 3937-3946.
[20]	Jiang K, Wang Z, Yi P, et al. Multi-scale progressive fusion network for single image deraining[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, USA,2020: 8346-8355.
[21]	Kupyn O, Budzan V, Mykhailych M, et al. Deblurgan: blind motion deblurring using conditional adversarial networks[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,Salt Lake City, USA, 2018: 8183-8192.
[22]	Isola P, Zhu J Y, Zhou T, et al. Image-to-image translation with conditional adversarial networks[C]∥Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition,Honolulu, USA, 2017: 1125-1134.
[23]	Ye Y, Chang Y, Zhou H, et al. Closing the loop: joint rain generation and removal via disentangled image translation[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, USA, 2021: 2053-2062.
[24]	Huang H, Yu A, He R. Memory oriented transfer learning for semi-supervised image deraining[C]∥Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern recognition,Nashville, USA, 2021: 7732-7741.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

数据集	Rain100L		Rain100H		Rain400
数据集	PSNR	SSIM	PSNR	SSIM	PSNR	SSIM
DID-MDN^［17］	26.84	0.860	25.00	0.754	25.01	0.728
DDN^［15］	32.38	0.926	24.95	0.781	25.05	0.721
DerainCycleGAN^［13］	31.49	0.936	26.82	0.848	29.24	0.852
RESCAN^［18］	34.99	0.931	26.45	0.846	27.14	0.841
JORDER^［16］	35.11	0.951	22.15	0.674	27.78	0.867
MSPFN^［20］	34.40	0.943	27.62	0.860	29.18	0.896
ID-CGAN^［11］	32.44	0.950	24.16	0.747	29.87	0.898
PRENET^［19］	34.79	0.945	28.06	0.888	29.92	0.902
JRGR^［23］	32.80	0.938	27.40	0.867	31.88	0.938
MOEDN^［24］	34.39	0.945	28.00	0.876	30.12	0.919
本文	35.50	0.962	28.98	0.897	32.32	0.946

评估指标	N=2	N=3	N=4	N=5
PSNR	35.02	35.28	35.50	35.11
SSIM	0.941	0.958	0.962	0.961

基于渐进递归的生成对抗单幅图像去雨算法

Progressive recursive generative adversarial network-based single-image rain removal algorithm

RICH HTML

PDF (PC)

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 24

相关文章 15

Metrics

本文评价

推荐阅读 10

模块	PSNR	SSIM
渐进递归模块	33.19	0.902
多尺度特征模块	34.74	0.918
渐进递归模块+多尺度特征模块	35.50	0.962

[1]	潘义勇,徐家聪,尤逸文,全勇俊. 网约车出行需求影响因素多尺度空间异质性分析[J]. 吉林大学学报(工学版), 2025, 55(5): 1567-1575.
[2]	侯越,郭劲松,林伟,张迪,武月,张鑫. 分割可跨越车道分界线的多视角视频车速提取方法[J]. 吉林大学学报(工学版), 2025, 55(5): 1692-1704.
[3]	才华,王玉瑶,付强,马智勇,王伟刚,张晨洁. 基于注意力机制和特征融合的语义分割网络[J]. 吉林大学学报(工学版), 2025, 55(4): 1384-1395.
[4]	刘广文,谢欣月,付强,才华,王伟刚,马智勇. 基于时空模板焦点注意的Transformer目标跟踪算法[J]. 吉林大学学报(工学版), 2025, 55(3): 1037-1049.
[5]	李学军,权林霏,刘冬梅,于树友. 基于Faster-RCNN改进的交通标志检测算法[J]. 吉林大学学报(工学版), 2025, 55(3): 938-946.
[6]	程德强,刘规,寇旗旗,张剑英,江鹤. 基于自适应大核注意力的轻量级图像超分辨率网络[J]. 吉林大学学报(工学版), 2025, 55(3): 1015-1027.
[7]	季渊,虞雅淇. 基于密集卷积生成对抗网络与关键帧的说话人脸视频生成优化算法[J]. 吉林大学学报(工学版), 2025, 55(3): 986-992.
[8]	才华,郑延阳,付强,王晟宇,王伟刚,马智勇. 基于多尺度候选融合与优化的三维目标检测算法[J]. 吉林大学学报(工学版), 2025, 55(2): 709-721.
[9]	张曦,库少平. 基于生成对抗网络的人脸超分辨率重建方法[J]. 吉林大学学报(工学版), 2025, 55(1): 333-338.
[10]	王宏志,宋明轩,程超,解东旋. 基于改进YOLOv5算法的道路目标检测方法[J]. 吉林大学学报(工学版), 2024, 54(9): 2658-2667.
[11]	朱圣杰,王宣,徐芳,彭佳琦,王远超. 机载广域遥感图像的尺度归一化目标检测方法[J]. 吉林大学学报(工学版), 2024, 54(8): 2329-2337.
[12]	赖丹晖,罗伟峰,袁旭东,邱子良. 复杂环境下多模态手势关键点特征提取算法[J]. 吉林大学学报(工学版), 2024, 54(8): 2288-2294.
[13]	才华,寇婷婷,杨依宁,马智勇,王伟刚,孙俊喜. 基于轨迹优化的三维车辆多目标跟踪[J]. 吉林大学学报(工学版), 2024, 54(8): 2338-2347.
[14]	温晓岳,钱国敏,孔桦桦,缪月洁,王殿海. TrafficPro：一种针对城市信控路网的路段速度预测框架[J]. 吉林大学学报(工学版), 2024, 54(8): 2214-2222.
[15]	郭昕刚,何颖晨,程超. 抗噪声的分步式图像超分辨率重构算法[J]. 吉林大学学报(工学版), 2024, 54(7): 2063-2071.