基于卷积神经网络的水下图像增强方法

doi:10.13229/j.cnki.jdxbgxb20170815

摘要/Abstract

摘要：

为了避免传统水下图像增强方法需要人工选取特征,特征提取困难、清晰度低的问题,本文提出一种基于卷积神经网络的水下图像增强方法。首先根据清晰图像,通过水下图像退化模型模拟水下图像形成过程并建立水下图像库。然后,提出一种用于增强水下图像的卷积神经网络模型,该模型可以直接在清晰图像和水下图像之间建立映射关系。最后,通过卷积神经网络提取的图像特征,对水下图像进行恢复。试验结果表明:本文方法较传统方法噪声更少,并且提高了清晰度,为水下图像增强方法的研究提供了新的思路。

关键词: 信息处理技术, 卷积神经网络, 水下图像, 图像增强, 清晰度

Abstract:

In order to avoid the problems of manually selecting features, difficulties in feature extraction and low visibility in traditional underwater image enhancing methods, an underwater image enhancing method based on convolution neural network was proposed. First, an underwater image dataset was established by simulation of underwater image formatting process by using clear images and underwater degraded model. Then, a convolution neural network model for underwater image enhancement was established, which can directly create a mapping relationship between the clear image and the underwater image. Finally, underwater images were restored by the features extracted by convolution neural network of underwater image. The experimental results show that the proposed method has less noise and improves contrast than tradition methods, and the algorithm provides a new idea for the underwater image enhancing research.

Key words: information procession technology, convolutional neural network, underwater images, image enhancing, visibility

中图分类号:

TP391.4

徐岩,孙美双. 基于卷积神经网络的水下图像增强方法[J]. 吉林大学学报(工学版), 2018, 48(6): 1895-1903.

XU Yan,SUN Mei-shuang. Enhancing underwater image based on convolutional neural networks[J]. Journal of Jilin University(Engineering and Technology Edition), 2018, 48(6): 1895-1903.

图/表 14

图1

图2

图3

图4

图5

表1

表2

表3

图6

图7

图8

表4

图9

表5

参考文献 26

[1]	Fattal R . Single image dehazing[J]. ACM Transactions on Graphics, 2008,27(3):11-15.
[2]	张凯, 裘溯, 王霞 , 等. 水下彩色图像的亮度通道多尺度Retinex增强算法[J]. 红外技术, 2011,33(11):630-634.
	Zhang Kai, Qiu Su, Wang Xia , et al. Multi-scale retinex enhancement algorithm on luminance channel of color underwater image[J]. Infrared Technology, 2011,33(11):630-634.
[3]	Land E H, McCann J . Lightness and Retinex theory[J]. Journal of the Optical Society of America, 1971,61(1):1-11. doi: 10.1364/JOSA.61.000001
[4]	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: 10.1109/TPAMI.2010.168 pmid: 20820075
[5]	Chiang J Y, Chen Y C . Underwater image ehancement by wavelength compensation and dehazing[J]. IEEE Transactions on Image Processing, 2012,21(4):1756-1769. doi: 10.1109/TIP.2011.2179666 pmid: 22180510
[6]	Ancuti C O, Bekaert P, Haber T, et al. Enhancing underwater images and videos by fusion [C]//2012 IEEE Conference on Computer Vision and Pattern Recognition, RI, USA, 2012: 81-88.
[7]	刘智, 黄江涛, 冯欣 . 构建多尺度深度卷积神经网络行为识别模型[J]. 光学精密工程, 2017,25(3):799-805. doi: 10.3788/OPE.20172503.0799
	Liu Zhi, Huang Jiang-tao, Feng Xin . Action recognition model construction based on multi-scale deep convolution neural network[J]. Optics and Precision Engineering, 2017,25(3):799-805. doi: 10.3788/OPE.20172503.0799
[8]	熊昌镇, 单艳梅, 郭芬红 . 结合主体检测的图像检索方法[J]. 光学精密工程, 2017,25(3):792-798. doi: 10.3788/OPE.20172503.0792
	Xiong Chang-zhen, Shan Yan-mei, Guo Fen-hong . Image retrieval method based on image principal part detection[J]. Optics and Precision Engineering, 2017,25(3):792-798. doi: 10.3788/OPE.20172503.0792
[9]	李琳辉, 伦智梅, 连静 , 等. 基于卷积神经网络的道路车辆检测方法[J]. 吉林大学学报:工学版, 2017,47(2):384-391. doi: 10.13229/j.cnki.jdxbgxb201702006
	Li Lin-hui, Lun Zhi-mei, Lian Jing , et al. Convolution neural network-based vehicle detection method[J]. Journal of Jilin University (Engineering and Technology Edition), 2017,47(2):384-391. doi: 10.13229/j.cnki.jdxbgxb201702006
[10]	Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation [C]//2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, USA, 2014: 580-587.
[11]	Liu S, Liang X, Liu L, et al. Matching-CNN meets KNN: quasi-parametric human parsing [C]//2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 2015: 1419-1427.
[12]	Cai B, Xu X, 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: 10.1109/TIP.2016.2598681
[13]	Jaffe J S . Computer modeling and the design of optimal underwater imaging systems[J]. IEEE Journal of Oceanic Engineering, 2002,15(2):101-111. doi: 10.1109/48.50695
[14]	杨爱萍, 郑佳, 王建 , 等. 基于颜色失真去除与暗通道先验的水下图像复原[J]. 电子与信息学报, 2015,37(11):2541-2547. doi: 10.11999/JEIT150483
	Yang Ai-ping, Zheng Jia, Wang Jian , et al. Underwater image restoration based on color cast removal and dark channel prior[J]. Journal of Electronics and Information Technology, 2015,37(11):2541-2547. doi: 10.11999/JEIT150483
[15]	马淼, 李贻斌 . 基于多级图像序列和卷积神经网络的人体行为识别[J]. 吉林大学学报:工学版, 2017,47(4):1244-1252. doi: 10.13229/j.cnki.jdxbgxb201704033
	Ma Miao, Li Yi-bin . Multi-level image sequences and convolutional neural networks based human action recognition method[J]. Journal of Jilin University (Engineering and Technology Edition), 2017,47(4):1244-1252. doi: 10.13229/j.cnki.jdxbgxb201704033
[16]	刘万军, 梁雪剑, 曲海成 . 不同池化模型的卷积神经网络学习性能研究[J]. 中国图象图形学报, 2016,21(9):1178-1190.
	Liu Wan-jun, Liang Xue-jian, Qu Hai-cheng . Learning performance of convolutional neural networks with different pooling models[J]. Journal of Image and Graphics, 2016,21(9):1178-1190.
[17]	Duntley S Q . Light in the sea[J]. Journal of the Optical Society of America, 1963,53(2):214-233. doi: 10.1364/JOSA.53.000214
[18]	默顿斯 L E, 张闻迪 . 水中摄影学[M]. 北京: 科学出版社, 1979.
[19]	Lecun Y, Bottou L, Bengio Y , et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998,86(11):2278-2324. doi: 10.1109/5.726791
[20]	Scharstein D, Szeliski R, Zabih R . A taxonomy and evaluation of dense two-frame stereo correspondence algorithms[J]. International Journal of Computer Vision, 2002,47(1-3):7-42. doi: 10.1023/A:1014573219977
[21]	Scharstein D, Szeliski R. High-accuracy stereo depth maps using structured light [C]//2003 IEEE Computer Society Conference on Computer Vision and Pattern Recognition,Madison, WI, USA, 2003: 195-202.
[22]	Scharstein D, Pal C . Learning conditional random Fields for stereo[J/OL]. [2017-07-30]..
[23]	Hirschmuller H, Scharstein D . Evaluation of cost functions for stereo matching[J/OL].[2017-08-01]..
[24]	Dong C, Loy C C, He K M , et al. Image super-resolution using deep convolutional networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2016,38(2):295-307. doi: 10.1109/TPAMI.2015.2439281 pmid: 26761735
[25]	Kwon H H, Tai Y W, Lin S. Data-driven depth map refinement via multi-scale sparse representation [C]//2015 IEEE Conference on Computer Vision and Pattern Recognition, Boston, MA, USA, 2015: 159-167.
[26]	Fu X Y, Zhuang P X, Huang Y, et al. A retinex-based enhancing approach for single underwater image [C]//2014 IEEE International Conference on Image Processing, Paris, France, 2014: 4572-4576.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed

试验编号	第1层卷积核尺寸	第2层卷积核尺寸	第3层卷积核尺寸	第4层卷积核尺寸	平均PSNR/dB
1	7×7	5×5	5×5	3×3	26.363
2	5×5	5×5	5×5	3×3	26.181
3	5×5	5×5	3×3	3×3	26.121
4	3×3	3×3	3×3	3×3	26.080

层的名称	特征图个数	卷积核尺寸	特征图扩充尺寸	卷积核移动步长
1	64	7×7	2	1
2	32	5×5	2	1
3	16	5×5	2	1
4	3	3×3	2	1

指标	算法	图片
指标	算法	Teddy	Art	Reindeer	Aloe	Dolls
PSNR/dB	Fattal	10.134	11.759	13.001	10.012	10.681
	Fu	15.712	14.251	12.136	11.709	15.024
	Ancuti	14.891	11.971	10.484	16.094	12.946
	本文	25.926	25.104	25.585	26.801	26.599
RMSE	Fattal	8.639	7.687	7.039	8.839	8.209
	Fu	5.953	6.489	7.046	7.477	6.061
	Ancuti	6.179	7.274	7.851	5.771	6.778
	本文	3.804	3.058	3.141	2.909	3.048

算法	均值	标准差	信息熵	平均梯度
Fattal	98.4971	61.3326	6.8398	6.2389
Fu	135.9848	69.0856	7.0569	8.4456
Ancuti	144.8726	68.5824	7.7185	9.1866
本文	143.2409	69.2448	7.7712	9.5125