渐进式时空细节增强视频去雾算法

doi:10.11996/JG.j.2095-302X.2026030576

摘要/Abstract

摘要：

针对现有视频去雾方法在纹理细节恢复不足及帧间一致性建模不充分的问题，提出了一种渐进式时空细节增强视频去雾算法(PSTD-net)，包括空间细节增强与跨帧时序建模2个模块。首先，为缓解现有方法对高频纹理建模能力不足的问题，提出了渐进式细节增强编码器，有效提取并恢复被雾霾破坏的纹理细节；其次，为提升视频帧间的时间一致性，设计了跨帧细节增强模块，通过时空注意力机制建模帧间依赖关系，增强图像细节的同时抑制伪影生成。实验在多个合成及真实雾霾视频数据集上进行，结果表明，PSTD-net在去雾效果和视觉一致性方面均取得了改进，为视频去雾任务提供了新的解决方案。

关键词: 视频去雾, 渐进式细节增强, 时间注意力, 差分卷积, 视频修复

Abstract:

To address insufficient texture detail restoration and inadequate inter-frame consistency modeling in existing video dehazing methods, a progressive spatiotemporal detail-enhanced video dehazing algorithm (Progressive Spatiotemporal Detail Enhancement Network，PSTD-net) was proposed, comprising two modules: a spatial detail enhancement and a cross-frame temporal modeling. First, to alleviate insufficient high-frequency texture modeling in existing methods, a progressive detail-enhancement encoder was proposed to effectively extract and restore texture details damaged by haze. Second, to improve the temporal consistency between video frames, a cross-frame detai- enhancement module was designed to model inter-frame dependencies through a spatiotemporal attention mechanism, enhancing image details while suppressing artifact generation. Experiments conducted on multiple synthetic and real hazy video datasets showed that PSTD-net improved both dehazing effect and visual consistency, providing a new solution for video dehazing tasks.

Key words: video dehazing, progressive detail enhancement, temporal attention, differential convolution, video restoration

中图分类号:

张益, 王振, 刘艳丽, 邢冠宇. 渐进式时空细节增强视频去雾算法[J]. 图学学报, 2026, 47(3): 576-588.

ZHANG Yi, WANG Zhen, LIU Yanli, XING Guanyu. A progressive spatiotemporal detail enhancement algorithm for video dehazing[J]. Journal of Graphics, 2026, 47(3): 576-588.

图/表 12

图1 PSTD-net的总体结构图

Fig. 1 Overall structure of PSTD-net

图2 跨帧细节增强模块的结构示意图

Fig. 2 Architecture of the cross-frame detail enhancement module

图3 多尺度邻域RWKV模块的结构示意图

Fig. 3 Architecture of the multi-scale neighborhood RWKV module

表1 各方法在HazeWorld和REVIDE上的定量指标对比

Table 1 Quantitative results on HazeWorld and REVIDE

方法	年份	类型	HazeWorld		REVIDE
方法	年份	类型	PSNR/dB	SSIM	PSNR/dB	SSIM
DCP^[3]	2011	图像去雾	16.49	0.812 6	11.03	0.728 5
Griddehazenet^[6]	2019	图像去雾	22.80	0.921 7	19.69	0.854 5
FFA^[41]	2020	图像去雾	22.11	0.900 7	16.65	0.813 3
MSDBN^[42]	2020	图像去雾	23.70	0.885 8	22.01	0.875 9
Dehamer^[43]	2022	图像去雾	22.92	0.904 4	19.43	0.700 8
Dehazeformer^[17]	2023	图像去雾	25.44	0.928 6	19.92	0.814 9
DEA-Net^[29]	2024	图像去雾	28.29	0.918 8	20.28	0.778 8
VDH^[8]	2018	视频去雾	17.97	0.778 0	16.64	0.813 3
EDVR^[44]	2019	视频去雾	22.91	0.903 6	21.22	0.870 7
CG-IDN^[10]	2021	视频去雾	25.25	0.915 5	23.21	0.883 6
BasicVSR++^[45]	2022	视频去雾	26.06	0.920 7	21.68	0.872 6
MAP-Net^[11]	2023	视频去雾	27.12	0.934 9	24.03	0.892 7
DVD-Net^[46]	2024	视频去雾	28.33	0.935 6	24.34	0.892 1
PSTD-net	—	视频去雾	28.66	0.938 7	23.65	0.896 7

图4 各方法在自然场景下对天空区域的修复效果对比

Fig. 4 Qualitative comparison of restoration performance on sky regions under natural scenarios ((a) Input image; (b) EDVR; (c) Dehamer; (d) Dehazeformer; (e) DEA-net; (f) MAP-Net; (g) PSTD-net; (h) Ground truth)

图5 各方法在自然场景下对天空区域的修复效果对比

Fig. 5 Qualitative comparison of sky region restoration in natural scenarios ((a) Input image; (b) EDVR; (c) Dehamer; (d) Dehazeformer; (e) DEA-net; (f) MAP-Net; (g) PSTD-net; (h) Ground truth)

图6 各方法在REVIDE数据集上的去雾效果对比

Fig. 6 Qualitative comparison of dehazing performance on the REVIDE dataset ((a) Input image; (b) EDVR; (c) Dehamer; (d) Dehazeformer; (e) DEA-net; (f) MAP-Net; (g) PSTD-net; (h) Ground truth)

表2 空间细节增强模块消融实验结果

Table 2 Results of ablation experiments for the spatial detail enhancement module

模型	PSNR/dB	SSIM	FLOPS/G	参数量/M
Base	26.69	0.882 4	63.43	6.81
Base-SDE	27.32	0.911 0	63.12	6.80
PSTD-net	27.61	0.917 2	63.42	6.81

图7 空间细节增强模块对去雾结果的影响

Fig. 7 Influence of the spatial detail enhancement module on dehazing performance ((a) Base; (b) Base-SDE; (c) PSTD-net; (d) Ground truth)

表3 跨帧细节增强模块消融实验结果

Table 3 Results of ablation experiments for the cross-frame detail enhancement module

模型	PSNR/dB	SSIM	FLOPS/G	参数量/ M
Base-Stack	26.53	0.8768	57.96	6.45
Base-SDE	27.43	0.9135	58.49	6.44
PSTD-net	27.61	0.9172	63.42	6.81

图8 空间细节增强模块对去雾结果的影响

Fig. 8 Influence of the spatial detail enhancement module on dehazing performance ((a) Base-satck; (b) Base-nodetail; (c) PSTD-net; (d) Ground truth)

表4 损失函数消融实验结果

Table 4 Results of ablation experiments for different loss terms

损失	PSNR	SSIM	迭代次数/ K
L₁	26.57	0.893 9	40
L₂	26.02	0.879 3	40
L₁+L₂	26.86	0.903 2	40
L₁+L_CR	27.61	0.917 2	40

参考文献 46

[1]	MCCARTNEY E J. Optics of the atmosphere: scattering by molecules and particles[M]. New York: John Wiley & Sons, 1976: 176-215.
[2]	TAN R T. Visibility in bad weather from a single image[C]// 2008 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2008: 1-8.
[3]	HE K M, SUN J, TANG X O. Single image haze removal using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(12): 2341-2353. DOI PMID
[4]	FATTAL R. Single image dehazing[J]. ACM Transactions on Graphics, 2008, 27(3): 1-9.
[5]	PANG Y W, XIE J, LI X L. Visual haze removal by a unified generative adversarial network[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2019, 29(11): 3211-3221. DOI URL
[6]	LIU X H, MA Y R, SHI Z H, et al. GridDehazeNet: attention-based multi-scale network for image dehazing[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 7313-7322.
[7]	CAI B L, XU X M, JIA K, et al. DehazeNet: an end-to-end system for single image haze removal[J]. IEEE Transactions on Image Processing, 2016, 25(11): 5187-5198. DOI PMID
[8]	REN W Q, ZHANG J G, XU X Y, et al. Deep video dehazing with semantic segmentation[J]. IEEE Transactions on Image Processing, 2019, 28(4): 1895-1908. DOI URL
[9]	LI B Y, PENG X L, WANG Z Y, et al. End-to-end united video dehazing and detection[C]//The 32nd AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2018: 859.
[10]	ZHANG X Y, DONG H, PAN J S, et al. Learning to restore hazy video: a new real-world dataset and a new method[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 9235-9244.
[11]	XU J Q, HU X W, ZHU L, et al. Video dehazing via a multi-range temporal alignment network with physical prior[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 18053-18062.
[12]	BERMAN D, TREIBITZ T, AVIDAN S. Non-local image dehazing[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2016: 1674-1682.
[13]	ZHENG Z R, REN W Q, CAO X C, et al. Ultra-high-definition image dehazing via multi-guided bilateral learning[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 16180-16189.
[14]	REN W Q, MA L, ZHANG J W, et al. Gated fusion network for single image dehazing[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 3253-3261.
[15]	ZHOU Y, CHEN Z H, LI P, et al. FSAD-Net: feedback spatial attention dehazing network[J]. IEEE Transactions on Neural Networks and Learning Systems, 2023, 34(10): 7719-7733. DOI URL
[16]	LI B Y, PENG X L, WANG Z Y, et al. AOD-Net: all-in-one dehazing network[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 4780-4788.
[17]	SONG Y D, HE Z Q, QIAN H, et al. Vision transformers for single image dehazing[J]. IEEE Transactions on Image Processing, 2023, 32: 1927-1941. DOI URL
[18]	庞宇, 吴天次, 王元发, 等. 快速视频去雾改进算法的FPGA实现[J]. 计算机应用研究, 2024, 41(6): 1803-1807.
	PANG Y, WU T C, WANG Y F, et al. FPGA implementation of improved algorithm for fast video de-fogging[J]. Application Research of Computers, 2024, 41(6): 1803-1807(in Chinese).
[19]	姚婷婷, 梁越, 柳晓鸣, 等. 基于雾线先验的时空关联约束视频去雾算法[J]. 电子与信息学报, 2020, 42(11): 2796-2804.
	YAO T T, LIANG Y, LIU X M, et al. Video dehazing algorithm via haze-line prior with spatiotemporal correlation constraint[J]. Journal of Electronics & Information Technology, 2020, 42(11): 2796-2804(in Chinese).
[20]	VAN NGUYEN T, VIEN A G, LEE C. Real-time image and video dehazing based on multiscale guided filtering[J]. Multimedia Tools and Applications, 2022, 81(25): 36567-36584. DOI
[21]	ZHANG J W, LI L, ZHANG Y, et al. Video dehazing with spatial and temporal coherence[J]. The Visual Computer, 2011, 27(6/8): 749-757. DOI URL
[22]	KIM J H, JANG W D, SIM J Y, et al. Optimized contrast enhancement for real-time image and video dehazing[J]. Journal of Visual Communication and Image Representation, 2013, 24(3): 410-425. DOI URL
[23]	SHAIKH J A H, MUKANE S M, RANDIVE S N. Lightweight progressive recurrent network for video de-hazing in adverse weather conditions[J]. The Visual Computer, 2025, 41(7): 4661-4672. DOI
[24]	LIU Y, WAN L, FU H Z, et al. Phase-based memory network for video dehazing[C]// The 30th ACM International Conference on Multimedia. New York: ACM, 2022: 5427-5435.
[25]	张万绪, 袁永德, 闫阳, 等. 基于暗原色先验的快速视频去雾优化算法[J]. 西北大学学报(自然科学版), 2016, 46(1): 43-47.
	ZHANG W X, YUAN Y D, YAN Y, et al. Fasting video haze removal algorithm using dark channel prior[J]. Journal of Northwest University (Natural Science Edition), 2016, 46(1): 43-47(in Chinese).
[26]	SHENG B, LI P, ALI R, et al. Improving video temporal consistency via broad learning system[J]. IEEE Transactions on Cybernetics, 2022, 52(7): 6662-6675. DOI URL
[27]	ZHU J, ZHANG Q W, FEI L K, et al. FFFN: frame-by-frame feedback fusion network for video super-resolution[J]. IEEE Transactions on Multimedia, 2023, 25: 6821-6835. DOI URL
[28]	WOO S, DEBNATH S, HU R H, et al. ConvNeXt V2: co-designing and scaling ConvNets with masked autoencoders[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2023: 16133-16142.
[29]	CHEN Z X, HE Z W, LU Z M. DEA-Net: single image dehazing based on detail-enhanced convolution and content-guided attention[J]. IEEE Transactions on Image Processing, 2024, 33: 1002-1015. DOI PMID
[30]	SU Z, LIU W Z, YU Z T, et al. Pixel difference networks for efficient edge detection[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 5097-5107.
[31]	YU Z T, ZHAO C X, WANG Z Z, et al. Searching central difference convolutional networks for face anti-spoofing[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 5294-5304.
[32]	PENG B, ALCAIDE E, ANTHONY Q, et al. RWKV: reinventing RNNs for the transformer era[EB/OL]. [2025-08-27]. https://aclanthology.org/2023.findings-emnlp.936/.
[33]	DUAN Y C, WANG W Y, CHEN Z, et al. efficient and scalable visual perception with RWKV-like architectures[EB/OL]. [2025-03-04]. https://iclr.cc/virtual/2025/poster/28412.
[34]	LI X, WANG W H, HU X L, et al. Selective kernel networks[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2019: 510-519.
[35]	WU H Y, QU Y Y, LIN S H, et al. Contrastive learning for compact single image dehazing[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 10546-10555.
[36]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2025-04-04]. https://dblp.org/rec/journals/corr/SimonyanZ14a.html.
[37]	LUCAS A, LOPEZ-TAPIA S, MOLINA R, et al. Generative adversarial networks and perceptual losses for video super-resolution[J]. IEEE Transactions on Image Processing, 2019, 28(7): 3312-3327. DOI PMID
[38]	WANG Z, BOVIK A C, SHEIKH H R, et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612. DOI PMID
[39]	PASZKE A, GROSS S, MASSA F, et al. PyTorch:an imperative style, high-performance deep learning library[C]// The 33rd International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2019: 721.
[40]	KINGMA D P, BA J. Adam: a method for stochastic optimization[EB/OL]. [2025-04-22]. https://dblp.org/rec/journals/corr/KingmaB14.html.
[41]	QIN X, WANG Z L, BAI Y C, et al. FFA-Net: feature fusion attention network for single image dehazing[C]// The 34th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2020: 11908-11915.
[42]	DONG H, PAN J S, XIANG L, et al. Multi-scale boosted dehazing network with dense feature fusion[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 2154-2164.
[43]	GUO C L, YAN Q X, ANWAR S, et al. Image dehazing transformer with transmission-aware 3D position embedding[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 5802-5810.
[44]	WANG X T, CHAN K C K, YU K, et al. EDVR: video restoration with enhanced deformable convolutional networks[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2019: 1954-1963.
[45]	CHAN K C K, ZHOU S C, XU X Y, et al. BasicVSR++: improving video super-resolution with enhanced propagation and alignment[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 5962-5971.
[46]	FAN J K, WENG J W, WANG K, et al. Driving-video dehazing with non-aligned regularization for safety assistance[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2024: 26109-26119.