基于多路混合注意力机制的水下图像增强网络

李云; 孙山林; 黄晴; 井佩光

doi:10.11999/JEIT230495

基于多路混合注意力机制的水下图像增强网络

doi: 10.11999/JEIT230495

李云¹,
孙山林³,
黄晴²,
井佩光^4, ,

1.
广西财经学院大数据与人工智能学院南宁 530003
2.
桂林电子科技大学信息与通信学院桂林 541000
3.
桂林航天工业学院电子信息与自动化学院桂林 541000
4.
天津大学电气自动化与信息工程学院天津 300072

基金项目: 国家自然科学基金(61861014)，博士启动基金(BS2021025)

详细信息

作者简介:
李云：女，教授，博士，研究方向为水下无线传感器网络、大数据分析和智能算法

孙山林：男，教授，博士，研究方向为图像处理和智能算法

黄晴：男，硕士生，研究方向为图像处理、人工智能

井佩光：男，副教授，博士，研究方向为多媒体信息处理、图像处理、人工智能

通讯作者:
井佩光　pgjing@tju.edu.cn

中图分类号: TN911.73
计量
- 文章访问数: 326
- HTML全文浏览量: 184
- PDF下载量: 98
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-26
- 修回日期: 2023-06-28
- 网络出版日期: 2023-07-03
- 刊出日期: 2024-01-17

Underwater Image Enhancement Network Based on Multi-channel Hybrid Attention Mechanism

1.
School of Big Data and Artificial Intelligence, Guangxi University of Finance and Economics, Nanning 530003, China
2.
School Of Information And Communication, Guilin University Of Electronic Technology, Guilin 541000, China
3.
School of Electronic Information and Automation, Guilin University of Aerospace Technology, Guilin 541000, China
4.
School of Electrical Automation and Information Engineering, Tianjin University, Tianjin 300072, China

Funds: The National Natural Science Foundation of China(61861014), Ph. D. Startup Fund(BS2021025)

摘要

摘要: 光线在水下被吸收或者散射使得水下图像成像出现色偏、模糊遮挡等问题，影响水下视觉任务。传统的图像增强方法分别采用直方图均衡、伽马矫正和白平衡方法较好地增强水下图像。然而，3种方法融合增强水下图像的互补性和相关性方面的研究较少。因此，该文提出一种基于多路混合注意力机制的水下图像增强网络。首先，提出多路特征提取模块，对图像进行直方图均衡支路、伽马矫正支路和白平衡支路的多路特征提取，提取图像的对比度、亮度和颜色特征；然后，融合直方图均衡、伽马矫正和白平衡3支路特征，增强3支路特征融合的互补性；最后，设计混合注意力学习模块，深度挖掘3支路在对比度、亮度和颜色的相关性矩阵，并引入跳跃连接增强图像输出。在多个数据集上的实验结果表明，该方法能够有效恢复水下图像色偏、模糊遮挡和提高图像明亮度。
- 水下图像增强 /
- 深度学习 /
- 注意力机制 /
- 跳跃连接
Abstract: The absorption or scattering of light under water causes problems such as color cast, blur and occlusion in underwater image imaging, which affects underwater vision tasks. Traditional image enhancement methods use histogram equalization, gamma correction and white balance methods to enhance underwater images well. However, there are few studies on the complementarity and correlation of the three methods fused to enhance underwater images. Therefore, an underwater image enhancement network based on multi-channel hybrid attention mechanism is proposed. Firstly, a multi-channel feature extraction module is proposed to extract the contrast, brightness and color features of the image by multi-channel feature extraction of histogram equalization branch, gamma correction branch and white balance branch. Then, the three branch features of histogram equalization, gamma correction and white balance are fused to enhance the complementarity of three branch feature fusion. Finally, a hybrid attention learning module is designed to deeply mine the correlation matrix of the three branches in contrast, brightness and color, and skip connections are introduced to enhance the image output. Experimental results on multiple datasets show that the proposed method can effectively recover the color cast, blur occlusion and improve the brightness of underwater images.
- Underwater image enhancement /
- Deep learning /
- Attention mechanism /
- Skip connection

HTML全文

图 1 网络结构图

下载: 全尺寸图片幻灯片

图 2 消融实验训练损失曲线图

下载: 全尺寸图片幻灯片

图 3 UIEB数据集上不同方法生成的水下图像视觉对比

下载: 全尺寸图片幻灯片

图 4 EUVP1数据集上不同方法生成的水下图像视觉对比

下载: 全尺寸图片幻灯片

图 5 EUVP2数据集上不同方法生成的水下图像视觉对比

下载: 全尺寸图片幻灯片

图 6 LSUI数据集上不同方法生成的水下图像视觉对比

下载: 全尺寸图片幻灯片

表 1 4种不同水下数据集的划分

	UIEB	EUVP1	EUVP2	LSUI
总图像	890	3700	2185	4279
训练集	790	3330	1966	3879
测试集	100	370	219	400

下载: 导出CSV

表 2 网络结构消融实验数值结果

	MSE(×10³)	PSNR	SSIM
A	0.9329	19.5549	0.8759
A+B	4.6547	11.8424	0.6027
A+C	1.0886	18.6155	0.8698
A+D	11.6075	7.6427	0.1967
A+B+C	0.9096	19.7709	0.8864
A+B+D	0.7690	20.2661	0.8901
A+C+D	0.8062	20.3661	0.8925
A+B+C+D	0.7671	20.3885	0.8798
A+B+C+D+E	0.7193	20.7678	0.8958

下载: 导出CSV

表 3 UIEB数据集上不同方法的定量评价

	MSE(×10³)	PSNR	SSIM	UCIQE	运行时间(s)
原图	1.5160	18.3782	0.7763	0.3934	–
Fusion	0.9764	28.1235	0.8428	0.4832	67.85
GDCP	3.6173	13.5613	0.7365	0.4560	579.41
Dehaze	2.3981	15.7978	0.7544	0.4624	12.74
HLRP	4.6692	12.1880	0.2309	0.4673	300.11
BRUE	2.6633	15.0887	0.6555	0.4257	249.12
Shallow-UWnet	1.2263	18.4789	0.8041	0.3566	30.48
WaterNet	1.0258	18.7704	0.8777	0.4165	64.85
UResnet	1.4343	17.9562	0.7615	0.4065	34.74
本文	0.7193	20.7678	0.8958	0.4076	57.85

下载: 导出CSV

表 4 EUVP1数据集上不同方法的定量评价

	MSE(×10³)	PSNR	SSIM	UCIQE	运行时间(s)
原图	1.5162	16.9408	0.7282	0.4244	–
Fusion	1.5543	16.5457	0.6606	0.4711	30.87
GDCP	4.3538	12.4384	0.5922	0.4735	223.98
Dehaze	2.9950	14.0605	0.6422	0.4575	7.87
HLRP	4.1953	12.2690	0.1295	0.4607	144.62
BRUE	1.9737	16.3921	0.6504	0.4281	85.99
Shallow-UWnet	0.3945	22.7718	0.8085	0.3938	5.28
WaterNet	0.4133	22.8706	0.8339	0.4128	6.90
UResnet	0.3237	23.7804	0.8296	0.4370	8.81
本文	0.3651	23.3899	0.8309	0.4378	7.49

下载: 导出CSV

表 5 EUVP2数据集上不同方法的定量评价

	MSE(×10³)	PSNR	SSIM	UCIQE	运行时间(s)
原图	0.7644	19.8464	0.7361	0.4274	–
Fusion	1.5287	16.8524	0.6682	0.4822	19.44
GDCP	3.2112	13.7086	0.6227	0.4794	149.14
Dehaze	2.0752	16.2689	0.6705	0.4646	5.25
HLRP	4.0604	12.3013	0.1669	0.4718	95.323
BRUE	2.0339	15.9520	0.6626	0.4324	58.81
Shallow-UWnet	0.3234	23.5612	0.7780	0.4128	4.71
WaterNet	0.4649	22.3487	0.7997	0.4207	10.00
UResnet	0.2982	24.0810	0.8034	0.4314	10.11
本文	0.3732	23.1460	0.8055	0.4063	5.40

下载: 导出CSV

表 6 LSUI数据集上不同方法的定量评价

	MSE(×10³)	PSNR	SSIM	UCIQE	运行时间(s)
原图	1.0684	18.8053	0.7977	0.4194	–
Fusion	1.3242	17.6604	0.7277	0.4825	34.72
GDCP	3.5459	13.4069	0.6766	0.4796	363.02
Dehaze	2.3828	15.4675	0.7192	0.4674	9.64
HLRP	3.8949	12.5132	0.1918	0.4915	240.86
BRUE	1.7558	16.6739	0.7074	0.4301	162.92
Shallow-UWnet	0.8527	19.6673	0.8046	0.3709	9.24
WaterNet	0.4931	22.1336	0.8628	0.4370	13.18
UResnet	0.5831	21.1576	0.8124	0.4243	13.13
本文	0.5576	21.5563	0.8552	0.4004	15.50

下载: 导出CSV

表 7 不同方法的运行时间和FPS

	Fusion	GDCP	Dehaze	HLRP	BRUE	Shallow-UWnet	WaterNet	UResnet	本文
时间(s)	67.85	579.41	12.74	300.11	249.12	30.48	64.85	34.74	57.85
Fps	1.47	0.17	7.85	0.33	0.40	3.28	1.54	2.88	1.73

下载: 导出CSV

参考文献(28)

[1]	侯冬, 任军委, 郭广坤, 等. 高精度水下激光频率传递研究进展[J]. 光电工程, 2023, 50(2): 220149. doi: 10.12086/oee.2023.220149 HOU Dong, REN Junwei, GUO Guangkun, et al. Progress on high-precision laser-based underwater frequency transfer[J]. Opto-Electronic Engineering, 2023, 50(2): 220149. doi: 10.12086/oee.2023.220149
[2]	PIZER S M, AMBURN E P, AUSTIN J D, et al. Adaptive histogram equalization and its variations[J]. Computer Vision, Graphics, and Image Processing, 1987, 39(3): 355–368. doi: 10.1016/S0734-189X(87)80186-X
[3]	PIZER S M, JOHNSTON R E, ERICKSEN J P, et al. Contrast-limited adaptive histogram equalization: Speed and effectiveness[C]. The First Conference on Visualization in Biomedical Computing, Atlanta, USA, 1990: 337–345.
[4]	ANCUTI C, ANCUTI C O, HABER T, et al. Enhancing underwater images and videos by fusion[C]. 2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, USA, 2012: 81–88.
[5]	RIZZI A, GATTA C, and MARINI D. Color correction between gray world and white patch[C]. The SPIE 4662, Human Vision and Electronic Imaging VII, San Jose, USA, 2002: 1–10.
[6]	SINGH G, JAGGI N, VASAMSETTI S, et al. Underwater image/video enhancement using wavelet based color correction (WBCC) method[C]. 2015 IEEE Underwater Technology, Chennai, India, 2015: 1–5.
[7]	MUNIRAJ M and DHANDAPANI V. Underwater image enhancement by combining color constancy and dehazing based on depth estimation[J]. Neurocomputing, 2021, 460: 211–230. doi: 10.1016/j.neucom.2021.07.003
[8]	BERMAN D, LEVY D, AVIDAN S, et al. Underwater single image color restoration using haze-lines and a new quantitative dataset[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(8): 2822–2837. doi: 10.1109/TPAMI.2020.2977624
[9]	WANG Yi, LIU Hui, and CHAU L P. Single underwater image restoration using adaptive attenuation-curve prior[J]. IEEE Transactions on Circuits and Systems I:Regular Papers, 2018, 65(3): 992–1002. doi: 10.1109/TCSI.2017.2751671
[10]	PENG Y T, CAO Keming, and COSMAN P C. Generalization of the dark channel prior for single image restoration[J]. IEEE Transactions on Image Processing, 2018, 27(6): 2856–2868. doi: 10.1109/TIP.2018.2813092
[11]	LI Chongyi, ANWAR S, HOU Junhui, et al. Underwater image enhancement via medium transmission-guided multi-color space embedding[J]. IEEE Transactions on Image Processing, 2021, 30: 4985–5000. doi: 10.1109/TIP.2021.3076367
[12]	李钰, 杨道勇, 刘玲亚, 等. 利用生成对抗网络实现水下图像增强[J]. 上海交通大学学报, 2022, 56(2): 134–142. doi: 10.16183/j.cnki.jsjtu.2021.075 LI Yu, YANG Daoyong, LIU Lingya, et al. Underwater image enhancement based on generative adversarial networks[J]. Journal of Shanghai Jiaotong University, 2022, 56(2): 134–142. doi: 10.16183/j.cnki.jsjtu.2021.075
[13]	MARQUES T P and ALBU A B. L²UWE: A framework for the efficient enhancement of low-light underwater images using local contrast and multi-scale fusion[C]. The 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, Seattle, USA, 2020: 2286–2295.
[14]	LI Chongyi, GUO Chunle, REN Wenqi, et al. An underwater image enhancement benchmark dataset and beyond[J]. IEEE Transactions on Image Processing, 2020, 29: 4376–4389. doi: 10.1109/TIP.2019.2955241
[15]	SUN Xin, LIU Lipeng, LI Qiong, et al. Deep pixel-to-pixel network for underwater image enhancement and restoration[J]. IET Image Processing, 2018, 13(3): 469–474. doi: 10.1049/iet-ipr.2018.5237
[16]	方明, 刘小晗, 付飞蚺. 基于注意力的多尺度水下图像增强网络[J]. 电子与信息学报, 2021, 43(12): 3513–3521. doi: 10.11999/JEIT200836 FANG Ming, LIU Xiaohan, and FU Feiran. Multi-scale underwater image enhancement network based on attention mechanism[J]. Journal of Electronics &Information Technology, 2021, 43(12): 3513–3521. doi: 10.11999/JEIT200836
[17]	米泽田, 晋洁, 李圆圆, 等. 基于多尺度级联网络的水下图像增强方法[J]. 电子与信息学报, 2022, 44(10): 3353–3362. doi: 10.11999/JEIT220375 MI Zetian, JIN Jie, LI Yuanyuan, et al. Underwater image enhancement method based on multi-scale cascade network[J]. Journal of Electronics &Information Technology, 2022, 44(10): 3353–3362. doi: 10.11999/JEIT220375
[18]	ZHUANG Peixian, WU Jiamin, PORIKLI F, et al. Underwater image enhancement with hyper-laplacian reflectance priors[J]. IEEE Transactions on Image Processing, 2022, 31: 5442–5455. doi: 10.1109/TIP.2022.3196546
[19]	ZHUANG Peixian, LI Chongyi, and WU Jiamin. Bayesian retinex underwater image enhancement[J]. Engineering Applications of Artificial Intelligence, 2021, 101: 104171. doi: 10.1016/j.engappai.2021.104171
[20]	SZEGEDY C, VANHOUCKE V, IOFFE S, et al. Rethinking the inception architecture for computer vision[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, USA, 2016: 2818–2826.
[21]	BAHDANAU D, CHO K, and BENGIO Y. Neural machine translation by jointly learning to align and translate[C]. The 3rd International Conference on Learning Representations, San Diego, USA, 2014.
[22]	WANG Zhou, 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: 10.1109/TIP.2003.819861
[23]	ISLAM M J, XIA Youya, and SATTAR J. Fast underwater image enhancement for improved visual perception[J]. IEEE Robotics and Automation Letters, 2020, 5(2): 3227–3234. doi: 10.1109/LRA.2020.2974710
[24]	PENG Lintao, ZHU Chunli, and BIAN Liheng. U-shape transformer for underwater image enhancement[J]. IEEE Transactions on Image Processing, 2023, 32: 3066–3079. doi: 10.1109/TIP.2023.3276332
[25]	YANG Miao and SOWMYA A. An underwater color image quality evaluation metric[J]. IEEE Transactions on Image Processing, 2015, 24(12): 6062–6071. doi: 10.1109/TIP.2015.2491020
[26]	HE Kaiming, SUN Jian, and TANG Xiaoou. Single image haze removal using dark channel prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(12): 2341–2353. doi: 10.1109/TPAMI.2010.168
[27]	LIU Peng, WANG Guoyu, QI Hao, et al. Underwater image enhancement with a deep residual framework[J]. IEEE Access, 2019, 7: 94614–94629. doi: 10.1109/ACCESS.2019.2928976
[28]	NAIK A, SWARNAKAR A, and MITTAL K. Shallow-UWnet: Compressed model for underwater image enhancement (student abstract)[C]. The 35th AAAI Conference on Artificial Intelligence, 2021: 15853–15854.