用于夜视辅助驾驶的轻量化图像眩光去除方法

doi:10.11996/JG.j.2095-302X.2026010057

摘要/Abstract

摘要：

夜视环境下，强光源引发的眩光干扰显著降低图像质量，影响夜视辅助驾驶系统的感知性能，现有眩光去除算法面临鲁棒性不足、计算复杂度高以及光源信息丢失等问题。为此，提出了一种面向夜视辅助驾驶的轻量化图像眩光去除方法(NFR-Net+)旨在提升图像清晰度并满足移动端实时计算需求。首先设计特征过滤机制，结合残差连接策略，增强网络对复杂夜视场景的特征提取能力，有效抑制过拟合，从而在不同光照条件和眩光类型下实现稳定的眩光去除效果。其次，引入非线性无激活特征注意力模块，通过轻量化设计构建高效注意力机制，显著提升图像细节重建质量，同时将模型参数量降低约8.28%，运行内存减少约11.1%，大幅优化计算效率。此外，针对传统方法中光源信息过度去除导致图像自然度下降的问题，优化了分割网络中的光源提取模块，通过改进的光源分离策略，精确保留光源区域的亮度和纹理信息，确保输出图像的真实性和自然感。实验结果表明，NFR-Net+在结构相似性(SSIM)、峰值信噪比(PSNR)和学习感知图像块相似度(LPIPS)等图像质量评估指标上均优于现有主流方法，表现出更高的去眩光性能和细节保留能力。同时，该方法在多种夜视场景和不同硬件设备上均展现出良好的适应性，能够满足实时处理的高效性要求，为智能视觉系统在资源受限的移动端部署提供了可行性。进一步的消融实验验证了各模块的有效性，凸显了特征过滤和注意力机制在提升性能与降低资源消耗中的关键作用。且为夜间自动驾驶和智能监控等应用场景提供了高效、轻量化的解决方案。

关键词: 眩光去除, 夜视环境, 光源信息, 非线性无激活, 轻量化

Abstract:

In night-vision environments, image quality was significantly degraded by glare from intense light sources, impairing the performance of night-vision assisted driving systems. Existing flare-removal algorithms suffer from limited robustness, high computational complexity, and loss of light-source information. To address these challenges, a lightweight image flare-removal method, Night Flare Removal Network+ (NFR-Net+), was proposed to enhance image clarity while meeting the real-time computational demands of mobile devices. The approach first incorporated a feature-filtering mechanism combined with residual connection strategies to strengthen feature extraction capabilities, effectively mitigating overfitting and ensuring robust flare removal across diverse lighting conditions and flare types. Additionally, a nonlinear, activation-free feature attention module was introduced. Via a lightweight design, an efficient attention mechanism was constructed that significantly improved image-detail reconstruction while reducing model parameters by approximately 8.28% and runtime memory by about 11.1%, thereby optimizing computational efficiency. To tackle the issue of diminished image naturalness due to excessive light-source removal in traditional methods, an enhanced light-source extraction module was developed within the segmentation network. This module employed an improved light-source separation strategy to accurately preserve brightness and texture details in light-source regions, ensuring the authenticity and naturalness of output images. Experimental results demonstrated that NFR-Net+ surpassed state-of-the-art methods on image quality metrics such as Structural Similarity Index Measure (SSIM), Peak Signal-to-Noise Ratio (PSNR), and Learned Perceptual Image Patch Similarity (LPIPS), exhibiting superior flare-removal performance and detail preservation. The method also demonstrated strong adaptability across various night-vision scenarios and hardware devices, fulfilling the efficiency requirements for real-time processing. Ablation studies further validated the effectiveness of individual components, highlighting the critical role of feature filtering and attention mechanisms in balancing performance and resource consumption. This approach provided an efficient, lightweight solution for applications such as nighttime autonomous driving and intelligent surveillance.

Key words: flare removal, night vision environment, light source information, nonlinear no activation, lightweight

中图分类号:

TP391
U463

李晔, 贾俊洋, 黄冠, 李玉洁, 齐文婷, 刘岩. 用于夜视辅助驾驶的轻量化图像眩光去除方法[J]. 图学学报, 2026, 47(1): 57-67.

LI Ye, JIA Junyang, HUANG Guan, LI Yujie, QI Wenting, LIU Yan. A lightweight image flare removal method for night vision assisted driving[J]. Journal of Graphics, 2026, 47(1): 57-67.

图/表 11

图1 整体网络框架图

Fig. 1 Overall network framework diagram

图2 改进前后的局部增强窗口转换器块((a) 改进前；(b) 改进后)

Fig. 2 Locally enhanced window transformer block before and after improvement((a) Before improvement；(b) Improved)

图3 特征过滤机制

Fig. 3 Feature filtering mechanism

图4 简化的通道以及像素注意力机制

Fig. 4 Simplified channels and pixel attention mechanism((a) CA；(b) SCA；(c) PA；(d) SPA)

图5 光源提取模块的结构图

Fig. 5 The structure diagram of the light source extraction

图6 转换后的BDD100K数据集 ((a), (c) 处理前；(b), (d) 处理后)

Fig. 6 The transformed BDD100K dataset((a), (c) Before treatment；(b), (d) Processed)

图7 眩光合成方法

Fig. 7 Flare synthesis method

图8 本文与其他方法的实验结果比较((a) 输入；(b) 文献[37]；(c) 文献[38]；(d) 文献[13]；(e) 文献[20]；(f) 文献[16]；(g) 本文方法；(h) 真值)

Fig. 8 This paper compares the experimental results with those of other method ((a) Input; (b) Literature [37]; (c) Literature [38]; (d) Literature [13]; (e) Literature [20]; (f) Literature [16]; (g) Ours；(h) GT)

图9 NFR-Net+算法实验结果泛化示例((a) 输入；(b) 输出)

Fig. 9 Example of generalization of experimental results of NFR-Net+ algorithm ((a) Input; (b) Output)

表1 眩光去除算法性能比较

Table 1 Performance comparison of flare removal algorithms

方法	PSNR/dB↑	SSIM↑	LPIPS↓	推理时间/s↓
文献[37]	24.97	0.897	0.137	0.098 8
文献[38]	27.45	0.904	0.034	0.096 4
文献[13]	28.76	0.924	0.029	1.358 4
文献[20]	30.60	0.967	0.021	3.945 4
文献[16]	30.61	0.769	0.018	1.574 1
本文方法	30.68	0.973	0.020	0.076 1

表2 非线性无激活特征注意力模块消融实验结果

Table 2 Ablation experimental results of nonlinear activation free feature attention module

方法	运行内存/G↓	参数/百万↓	PSNR/dB↑
CA	10.14	1.65	30.53
SCA	9.67	1.61	30.58
PA	9.56	1.65	30.52
SPA	9.14	1.61	30.56
FA	10.63	2.07	30.60
NAFFA	9.75	1.84	30.65

参考文献 38

[1]	NUSSBERGER A, GRABNER H, VAN GOOL L. Robust aerial object tracking in images with lens flare[C]// 2015 IEEE International Conference on Robotics and Automation. New York: IEEE Press, 2015: 6380-6387.
[2]	YE S T, YIN J L, CHEN B H, et al. Single image glare removal using deep convolutional networks[C]// 2020 IEEE International Conference on Image Processing. New York: IEEE Press, 2020: 201-205.
[3]	WEN L H, JO K H. Deep learning-based perception systems for autonomous driving: a comprehensive survey[J]. Neurocomputing, 2022, 489: 255-270. DOI URL
[4]	BOYNTON P A, KELLEY E F. Liquid-filled camera for the measurement of high-contrast images[C]// The SPIE 5080, Cockpit Displays X. Bellingham: SPIE, 2003: 370-378.
[5]	RASKAR R, AGRAWAL A, WILSON C A, et al. Glare aware photography: 4D ray sampling for reducing glare effects of camera lenses[J]. ACM Transactions on Graphics, 2008, 27(3): 1-10.
[6]	MACLEOD H A. Thin-film optical filters[M]. Boca Raton: Taylor & Francis, 2010.
[7]	VITORIA P, BALLESTER C. Automatic flare spot artifact detection and removal in photographs[J]. Journal of Mathematical Imaging and Vision, 2019, 61(4): 515-533. DOI
[8]	ASHA C S, BHAT S K, NAYAK D, et al. Auto removal of bright spot from images captured against flashing light source[C]// 2019 IEEE International Conference on Distributed Computing, VLSI, Electrical Circuits and Robotics. New York: IEEE Press, 2019: 1-6.
[9]	时妙文, 范琳伟, 王桦, 等. 基于四元数组稀疏的彩色图像去噪[J]. 图学学报, 2023, 44(2): 298-303. DOI
	SHI M W, FAN L W, WANG H, et al. Quaternion patch-group sparse coding for color image denoising[J]. Journal of Graphics, 2023, 44(2): 298-303 (in Chinese).
[10]	徐祯东, 张天宇, 张世恒, 等. 基于YUV颜色空间GAN网络的图像去雾算法研究[J]. 图学学报, 2023, 44(5): 928-936. DOI
	XU Z D, ZHANG T Y, ZHANG S H, et al. Image defogging algorithm based on YUV color space GAN network[J]. Journal of Graphics, 2023, 44(5): 928-936 (in Chinese).
[11]	CHEN X, PAN J S, DONG J X. Bidirectional multi-scale implicit neural representations for image deraining[C]// 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2024: 25627-25636.
[12]	QIAN X T, HANCKE G P, LAU R W H. Light source guided single-image flare removal from unpaired data[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 4177-4185.
[13]	WU Y C, HE Q R, XUE T F, et al. How to train neural networks for flare removal[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 2239-2247.
[14]	RONNEBERGER O, FISCHER P, BROX T. U-Net: convolutional networks for biomedical image segmentation[C]// The 18th International Conference on Medical Image Computing and Computer-Assisted Intervention. Cham: Springer, 2015: 234-241.
[15]	DAI Y K, LI C Y, ZHOU S C, et al. Flare7K: a phenomenological nighttime flare removal dataset[C]// The 36th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2022: 284.
[16]	JIANG Y G, CHEN X H, PUN C M, et al. MFDNet: multi-Frequency Deflare Network for efficient nighttime flare removal[J]. The Visual Computer, 2024, 40(11): 7575-7588. DOI
[17]	LIANG J Y, CAO J Z, SUN G L, et al. SwinIR: image restoration using swin transformer[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2021: 1833-1844.
[18]	陈冠豪, 徐丹, 贺康建, 等. 基于转置注意力和CNN的图像超分辨率重建网络[J]. 图学学报, 2025, 46(1): 35-46. DOI
	CHEN G H, XU D, HE K J, et al. TSA-SFNet: transpose self-attention and CNN based stereoscopic fusion network for image super-resolution[J]. Journal of Graphics, 2025, 46(1): 35-46 (in Chinese). DOI
[19]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// The 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
[20]	WANG Z D, CUN X D, BAO J M, et al. Uformer: a general U-shaped transformer for image restoration[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 17662-17672.
[21]	SINHA D, EL-SHARKAWY M. Thin MobileNet: an enhanced MobileNet architecture[C]// 2019 IEEE 10th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference. New York: IEEE Press, 2019: 280-285.
[22]	KOONCE B. EfficientNet[M]// Convolutional Neural Networks with Swift for Tensorflow:Image Recognition and Dataset Categorization. Berkeley: Apress, 2021: 109-123.
[23]	FU X Y, LIANG B R, HUANG Y, et al. Lightweight pyramid networks for image deraining[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 31(6): 1794-1807. DOI PMID
[24]	LI B Y, PENG X L, WANG Z Y, et al. An all-in-one network for dehazing and beyond[EB/OL]. [2025-01-04]. https://arxiv.org/abs/1707.06543.
[25]	WANG T, ZHANG K H, SHEN T R, et al. Ultra-high- definition low-light image enhancement: a benchmark and transformer-based method[C]// The 37th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2023: 2654-2662.
[26]	YANG S Z, ZHANG X Y, WANG Y H, et al. DiffLLE: diffusion-based domain calibration for weak supervised low-light image enhancement[J]. International Journal of Computer Vision, 2025, 133(5): 2527-2546. DOI
[27]	SHEN Z, XU H Y, LUO T, et al. UDAformer: underwater image enhancement based on dual attention transformer[J]. Computers & Graphics, 2023, 111: 77-88. DOI URL
[28]	HU J, SHEN L, SUN G. Squeeze-and-excitation networks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 7132-7141.
[29]	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.
[30]	JIANG K, WANG Z Y, YI P, et al. Multi-scale progressive fusion network for single image deraining[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 8346-8355.
[31]	LIU X H, MA Y G, 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: 7314-7323.
[32]	YU F, KOLTUN V. Multi-scale context aggregation by dilated convolutions[EB/OL]. [2025-01-04]. ttps://dblp.uni-trier.de/db/conf/iclr/iclr2016.html#YuK15.
[33]	XU H L, LIU S H, SHU Y. Uformer++: light uformer for image restoration[C]// The 30th International Conference on Neural Information Processing. Cham: Springer, 2024: 365-376.
[34]	CHEN L Y, CHU X J, ZHANG X Y, et al. Simple baselines for image restoration[C]// The 17th European Conference on Computer Vision. Cham: Springer, 2022: 13-33.
[35]	SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. [2025-01-04]. https://dblp.uni-trier.de/db/conf/iclr/iclr2015.html#SimonyanZ14.
[36]	YU F, CHEN H F, WANG X, et al. BDD100K: a diverse driving dataset for heterogeneous multitask learning[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 2636-2645.
[37]	ZHANG J, CAO Y, ZHA Z J, et al. Nighttime dehazing with a synthetic benchmark[C]// The 28th ACM International Conference on Multimedia. New York: ACM, 2020: 2355-2363.
[38]	JIN Y Y, YANG W H, TAN R T. Unsupervised night image enhancement: when layer decomposition meets light-effects suppression[C]// The 17th European Conference on Computer Vision. Cham: Springer, 2022: 404-421.