基于两通道深度卷积神经网络的图像隐藏方法

段新涛; 王文鑫; 李磊; 邵志强; 王鲜芳; 秦川

doi:10.11999/JEIT210280

基于两通道深度卷积神经网络的图像隐藏方法

doi: 10.11999/JEIT210280

1.
河南师范大学计算机与信息工程学院新乡 453007
2.
河南工学院计算机科学与技术学院新乡 453003
3.
上海理工大学光电与计算机工程学院上海 200093

基金项目: 国家自然科学基金(U1904123, 61672354, 62072157), 教育人工智能与个性化学习河南省重点实验室基金

详细信息

作者简介:
段新涛：男，1972年生，副教授，研究方向为图像信息隐藏、图像盲取证、深度学习、盲源分离等

王文鑫：男，1994年生，硕士生，研究方向为图像信息隐藏、深度学习

李磊：男，1995年生，硕士生，研究方向为图像信息隐藏、深度学习

邵志强：男，1996年生，硕士生，研究方向为图像信息隐藏、深度学习

王鲜芳：女，1969年生，教授，研究方向为人工智能与模式识别、数据挖掘、机器学习及应用

秦川：男，1980年生，教授，研究方向为多媒体信息安全、数字图像处理、信息隐藏、AI安全、深度学习、密文域信号处理、数字取证

通讯作者:
段新涛　duanxintao@htu.edu.cn

中图分类号: TN911.73; TP309.2
计量
- 文章访问数: 958
- HTML全文浏览量: 289
- PDF下载量: 128
- 被引次数: 0
出版历程
- 收稿日期: 2021-04-06
- 修回日期: 2021-09-13
- 录用日期: 2021-09-13
- 网络出版日期: 2021-12-22
- 刊出日期: 2022-05-25

Image Hiding Method Based on Two-Channel Deep Convolutional Neural Network

1.
College of Computer and Information Engineering, Henan Normal University, Xinxiang 453007, China
2.
College of Computer Science and Technology, Henan Institute of Technology, Xinxiang 453003, China
3.
School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

Funds: The National Natural Science Foundation of China (U1904123, 61672354, 62072157), The Key Laboratory Foundation of Artificial Intelligence and Personalized Learning in Education of Henan Province

摘要

摘要: 现有的基于深度卷积神经网络(DCNN)实现的图像信息隐藏方法存在图像视觉质量差和隐藏容量低的问题。针对此类问题，该文提出一种基于两通道深度卷积神经网络的图像隐藏方法。首先，与以往的隐藏框架不同，该文提出的隐藏方法中包含1个隐藏网络和2个结构相同的提取网络，实现了在1幅载体图像上同时对2幅全尺寸秘密图像进行有效的隐藏和提取；其次，为了提高图像的视觉质量，在隐藏网络和提取网络中加入了改进的金字塔池化模块和预处理模块。在多个数据集上的测试结果表明，所提方法较现有的图像信息隐藏方法在视觉质量上有显著提升，载体图像PSNR和SSIM分别提高了3.75 dB和3.61%，实现的相对容量为2，同时具有良好的泛化能力。
- 图像信息隐藏 /
- 深度卷积神经网络 /
- 金字塔池化 /
- 预处理
Abstract: The existing image information hiding methods based on Deep Convolutional Neural Networks (DCNN) have the problems of poor image visual quality and low hiding capacity. Addressing such issues, an image hiding method based on a two-channel deep convolutional neural network is proposed. First, different from the previous hiding framework, the hiding method proposed in this paper includes one hiding network and two revealing networks with the same structure, and two full-size secret images can be effectively hidden and revealed at the same time is realized. Then, to improve the visual quality of the image, an improved pyramid pooling module and a preprocessing module are added to the hiding and revealing network. The test results on multiple data sets show that the proposed method has a significant improvement in visual quality compared with existing image information hiding methods. The Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity (SSIM) values are increased by 3.75 dB and 3.61 % respectively, a relative capacity of 2 and good generalization ability are achieved.
- Image information hiding /
- Deep Convolutional Neural Networks (DCNN) /
- Pyramid pooling /
- Preprocessing

HTML全文

图 1 隐藏框架

下载: 全尺寸图片幻灯片

图 2 改进模块

下载: 全尺寸图片幻灯片

图 3 隐藏网络和提取网络

下载: 全尺寸图片幻灯片

图 4 主观效果对比

下载: 全尺寸图片幻灯片

图 5 放大效果对比

下载: 全尺寸图片幻灯片

图 6 StegExpose隐写分析结果

下载: 全尺寸图片幻灯片

图 7 不同数据集的测试结果

下载: 全尺寸图片幻灯片

表 1 与文献[9]比较

图像	方法	载体图像-隐写图像		秘密图像1-提取图像1		秘密图像2-提取图像2
图像	方法	PSNR(dB)	SSIM(%)	PSNR(dB)	SSIM(%)	PSNR(dB)	SSIM(%)
图4(a)	文献[9]	31.95	94.44	29.68	83.90	26.93	78.43
图4(a)	本文方法	34.48	99.28	40.13	97.91	32.60	98.22
图4(b)	文献[9]	31.08	95.19	28.72	93.01	33.22	88.67
图4(b)	本文方法	38.17	98.47	37.16	98.11	34.35	97.28
平均值	文献[9]	32.32	94.81	30.40	90.70	30.55	90.29
平均值	本文方法	36.07	98.42	34.97	96.56	35.11	96.48

下载: 导出CSV

表 2 消融实验的PSNR和SSIM比较

图像	载体图像-隐写图像		秘密图像1-提取图像1		秘密图像2-提取图像2
图像	PSNR(dB)	SSIM(%)	PSNR(dB)	SSIM(%)	PSNR(dB)	SSIM(%)
ImageNet	36.07	98.42	34.97	96.56	35.11	96.48
*Prep	35.66	96.89	34.23	95.12	34.44	95.17
*Pyramid	34.86	95.65	33.85	94.93	34.10	96.27

下载: 导出CSV

表 3 隐写分析结果

隐藏模型	StegExpose AUC	SRNet 隐写分析准确率
本文方法	0.5533	0.6844
文献[9]	–	0.6975
*Prep	0.6223	0.7195
*Pyramid	0.5698	0.6994

下载: 导出CSV

表 4 嵌入容量比较

方法	绝对容量(Byte)	隐写图像大小(Byte)	相对容量
文献[21]	18.3～135.4	64$ \times $64	1.49×10^–3～1.10×10^–2
文献[22]	1535～4300	1024$ \times $1024	1.46×10^–3～4.10×10^–3
文献[23]	26214～104857	512$ \times $512	1×10^–1～4×10^–1
文献[10]	3$ \times $224$ \times $224	3$ \times $224$ \times $224	1
文献[12]	3$ \times $256$ \times $256～3$ \times $512$ \times $512	3$ \times $512$ \times $512	2.5×10^–1～1
本文方法	6$ \times $256$ \times $256	3$ \times $256$ \times $256	2

下载: 导出CSV

表 5 修改率和提取率比较(%)

图像	方法	载体图像修改率	秘密图像1提取率	秘密图像2提取率
图4(a)	文献[9]	1.35	98.65	97.44
图4(a)	本文方法	1.68	99.30	98.18
图4(b)	文献[9]	1.20	98.47	98.86
图4(b)	本文方法	1.09	98.97	98.47
ImageNet平均值	文献[9]	1.80	98.02	97.62
ImageNet平均值	本文方法	1.61	98.99	98.92

下载: 导出CSV

表 6 5组数据集的测试结果

数据集	载体图像-隐写图像			秘密图像1-提取图像1			秘密图像2-提取图像2
数据集	PSNR(dB)	SSIM(%)	修改率(%)	PSNR(dB)	SSIM(%)	提取率(%)	PSNR(dB)	SSIM(%)	提取率(%)
CeleA	35.81	96.40	1.83	36.18	96.90	98.22	36.24	97.50	98.24
COCO	34.03	96.41	2.22	33.29	93.79	97.62	33.82	94.98	97.76
VOC2012	34.06	96.48	2.23	33.45	93.54	97.66	34.10	94.80	97.80
AID	34.83	97.29	2.07	32.35	93.28	96.75	34.12	95.22	97.65
UCMerced Land Use	34.53	96.88	2.16	31.25	92.52	96.44	32.12	94.76	96.85

下载: 导出CSV

参考文献(23)

[1]	ZHANG Chaoning, LIN Chenguo, BENZ P, et al. A brief survey on deep learning based data hiding, steganography and watermarking[EB/OL]. https://arxiv.org/abs/2103.01607, 2021.
[2]	KER A D. Improved detection of LSB steganography in grayscale images[C]. The 6th International Workshop on Information Hiding (IH), Toronto, Canada, 2004: 97–115.
[3]	FILLER T, JUDAS J, and FRIDRICH J. Minimizing additive distortion in steganography using Syndrome-Trellis Codes[J]. IEEE Transactions on Information Forensics and Security, 2011, 6(3): 920–935. doi: 10.1109/TIFS.2011.2134094
[4]	FRIDRICH J and FILLER T. Practical methods for minimizing embedding impact in steganography[C]. SPIE 6505, Security, Steganography, and Watermarking of Multimedia Contents IX, San Jose, USA, 2007: 13–27.
[5]	FRIDRICH J, GOLJAN M, LISONEK P, et al. Writing on wet paper[C]. SPIE 5681, Security, Steganography, and Watermarking of Multimedia Contents VII, San Jose, USA, 2005: 328–340.
[6]	TANG Weixuan, LI Bin, BARNI M, et al. An automatic cost learning framework for image steganography using deep reinforcement learning[J]. IEEE Transactions on Information Forensics and Security, 2021, 16: 952–967. doi: 10.1109/TIFS.2020.3025438
[7]	ZHANG Chaoning, BENZ P, KARJAUV A, et al. UDH: Universal Deep Hiding for steganography, watermarking, and light field messaging[J]. Advances in Neural Information Processing Systems, 2020, 33: 10223–10234.
[8]	LUO Xiyang, ZHAN Ruohan, CHANG Huiwen, et al. Distortion agnostic deep watermarking[C]. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, USA, 2020: 13545–13554.
[9]	BALUJA S. Hiding images within images[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020, 42(7): 1685–1697. doi: 10.1109/TPAMI.2019.2901877
[10]	CHEN Feng, XING Qinghua, and LIU Fuxian. Technology of hiding and protecting the secret image based on two-channel deep hiding network[J]. IEEE Access, 2020, 8: 21966–21979. doi: 10.1109/ACCESS.2020.2969524
[11]	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 (CVPR), Las Vegas, USA, 2016: 2818–2826.
[12]	YU Chong. Attention based data hiding with generative adversarial networks[C]. The 34th AAAI Conference on Artificial Intelligence (AAAI), New York, USA, 2020: 1120–1128.
[13]	ZHU Junyan, PARK T, ISOLA P, et al. Unpaired image-to-image translation using cycle-consistent adversarial networks[C]. 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, 2017: 2242–2251.
[14]	ZHOU Bolei, KHOSLA A, LAPEDRIZA À, et al. Learning deep features for discriminative localization[C]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, USA, 2016: 2921–2929.
[15]	ZHOU Bolei, KHOSLA A, LAPEDRIZA À, et al. Object detectors emerge in deep scene CNNs[EB/OL]. http://arxiv.org/abs/1412.6856, 2015.
[16]	ZHAO Hengshuang, SHI Jianping, QI Xiaojuan, et al. Pyramid scene parsing network[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, USA, 2017: 6230–6239.
[17]	RONNEBERGER O, FISCHER P, and BROX T. U-Net: Convolutional networks for biomedical image segmentation[C]. The 18th International Conference on Medical Image Computing and Computer-Assisted Intervention, Munich, Germany, 2015: 234–241.
[18]	HORÉ A and ZIOU D. Image quality metrics: PSNR vs. SSIM[C]. The 20th International Conference on Pattern Recognition, Istanbul, Turkey, 2010: 2366–2369.
[19]	YE Yuanxin, SHAN Jie, BRUZZONE L, et al. Robust registration of multimodal remote sensing images based on structural similarity[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(5): 2941–2958. doi: 10.1109/TGRS.2017.2656380
[20]	BOEHM B. StegExpose - A tool for detecting LSB steganography[EB/OL]. http://arxiv.org/abs/1410.6656, 2014.
[21]	BOROUMAND M, CHEN Mo, and FRIDRICH J. Deep residual network for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2019, 14(5): 1181–1193. doi: 10.1109/TIFS.2018.2871749
[22]	WU Kuochen and WANG C. Steganography using reversible texture synthesis[J]. IEEE Transactions on Image Processing, 2015, 24(1): 130–139. doi: 10.1109/TIP.2014.2371246
[23]	YANG Jianhua, LIU Kai, KANG Xiangui, et al. Spatial image steganography based on generative adversarial network[OL]. http://arxiv.org/abs/1804.07939, 2018.