基于普遍对抗噪声的高效载体图像增强算法

doi:10.3969/j.issn.1671-1122.2022.02.008

摘要/Abstract

摘要：

在图像隐写领域,利用对抗样本技术实现载体图像增强是提升隐写安全性的重要手段。然而,现有基于对抗样本的载体图像增强算法需要对每张载体图像生成特定的对抗噪声,这导致算法时间效率低、实用性较差,并且增强后的载体图像对抗不同隐写分析器的迁移能力较弱。针对上述问题,文章提出一种基于普遍对抗噪声的高效载体图像增强算法,将对抗样本算法DeepFool作为基础,以对隐写分析器的攻击成功率为标准循环迭代构建普遍对抗噪声,将其与载体图像进行叠加完成载体图像增强,实现构建单个普遍对抗噪声便能对不同的载体图像进行增强。为进一步提高载体图像增强后抵抗不同隐写分析器的迁移能力,文章提出将针对不同隐写分析器的普遍对抗噪声进行融合得到复合普遍对抗噪声,实现对不同隐写分析器均具有良好隐写安全性的目的。实验结果表明,文章算法在载体图像增强的时间效率上相较于目前最先进的算法得到了极大程度提升,具有更强的实用性。此外,利用复合普遍对抗噪声进行载体图像增强能够显著提高不同隐写分析器的安全性。

关键词: 信息隐藏, 数字图像隐写术, 载体图像增强, 对抗样本

Abstract:

In the field of steganography, applying adversarial example technology to enhance cover image is an important method to improve the security of steganography. However, current enhancement methods of cover image based on adversarial example need to generate specific adversarial noise for each cover image independently, which leads to low efficiency and poor practicability. Furthermore, enhanced cover images have poor transfer ability against different steganalyzers. In order to solve the above-mentioned problems, this paper proposed an efficient enhancement algorithm of cover image based on universal adversarial noise. The proposed algorithm applied the adversarial example generation method DeepFool as the basic technique to construct a universal adversarial noise, which used a cyclic iterative strategy and assigned the success rate of attack on the steganalyzer as the target of optimization. Then the enhancement of cover image was achieved by adding universal adversarial noise to the cover image. For the proposed method, a single universal adversarial noise can enhance different cover images, which improves the time efficiency of cover enhancement significantly. Besides, to further improve the transfer ability of enhanced cover image, this paper fused the universal adversarial noise of different steganalyzers to obtain a fused universal adversarial noise. The use of fused universal adversarial noise can improve the security of the enhanced cover image to different steganalyzers. The experimental results show that the proposed method can achieve a significant improvement of time efficiency with other state-of-the-art algorithms. Futhermore, the cover image enhanced by the fused universal adversarial noise has much better security to different steganalyzers after embedding secret information.

Key words: information hiding, image steganography, enhancing cover image, adversarial example

中图分类号:

TP309

夏强, 何沛松, 罗杰, 刘嘉勇. 基于普遍对抗噪声的高效载体图像增强算法[J]. 信息网络安全, 2022, 22(2): 64-75.

XIA Qiang, HE Peisong, LUO Jie, LIU Jiayong. An Efficient Enhancement Algorithm of Cover Image Based on Universal Adversarial Noise[J]. Netinfo Security, 2022, 22(2): 64-75.

图/表 8

图1

图2

图3

表1

表2

表3

表4

表5

参考文献 21

[1]	XIONG Jinbo, MA Rong, ZHANG Yuanyuan, et al. Image Information Hiding Method and Implementation for Social Network[J]. Netinfo Security, 2017, 17(3):6-8.
	熊金波, 马蓉, 张媛媛, 等. 面向社交网络的图片信息隐藏方法与实现[J]. 信息网络安全, 2017, 17(3):6-8.
[2]	LI Dan, GAO Yong. Audio Information Hiding Algorithm Based on Time-frequency Modulation Method Applied in Broadcast Channel[J]. Journal of Sichuan University (Natural Science Edition), 2017, 54(2):298-302.
	李丹, 高勇. 适用于广播信道的时频调制音频隐藏算法[J]. 四川大学学报(自然科学版), 2017, 54(2):298-302.
[3]	HOLUB V, FRIDRICH J, DENEMARK T. Universal Distortion Function for Steganography in an Arbitrary Domain[J]. EURASIP Journal on Information Security, 2014(1):1-13.
[4]	DENEMARK T, FRIDRICH J, COMESAÑA-ALFARO P. Improving Selection-channel-aware Steganalysis Features[J]. Electronic Imaging, 2016(8):1-8.
[5]	XU Guanshuo, WU Hanzhou, SHI Yunqing. Structural Design of Convolutional Neural Networks for Steganalysis[J]. IEEE Signal Processing Letters, 2016, 23(5):708-712. doi: 10.1109/LSP.2016.2548421 URL
[6]	YE Jian, NI Jiangqun, YI Yang. Deep Learning Hierarchical Representations for Image Steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2017, 12(11):2545-2557. doi: 10.1109/TIFS.2017.2710946 URL
[7]	BOROUMAND M, CHEN M, FRIDRICH J. Deep Residual Network for Steganalysis of Digital Images[J]. IEEE Transactions on Information Forensics and Security, 2018, 14(5):1181-1193. doi: 10.1109/TIFS.2018.2871749 URL
[8]	SZEGEDY C, ZAREMBA W, SUTSKEVER I, et al. Intriguing Properties of Neural Networks[EB/OL]. https://arxiv.org/abs/1312.6199, 2014-02-19.
[9]	GOODFELLOW I J, SHLENS J, SZEGEDY C. Explaining and Harnessing Adversarial Examples[EB/OL]. https://arxiv.org/abs/1412.6572, 2015-03-20.
[10]	ZHANG Yiwei, ZHANG Weiming, CHEN Kejiang, et al. Adversarial Examples against Deep Neural Network Based Steganalysis[C]//ACM. Proceedings of the 6th ACM Workshop on Information Hiding and Multimedia Security, June 20-22, 2018, Innsbruck, Austria. New York: Association for Computing Machinery, 2018: 67-72.
[11]	MOOSAVI-DEZFOOLI S M, FAWZI A, FROSSARD P. Deepfool: A Simple and Accurate Method to Fool Deep Neural Networks[C]//IEEE. Proceedings of the 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 27-30, 2016, Las Vegas, NV, USA. New York: IEEE, 2016: 2574-2582.
[12]	HOLUB V, FRIDRICH J. Designing Steganographic Distortion Using Directional Filters[C]//IEEE. 2012 IEEE International Workshop on Information Forensics and Security (WIFS), December 2-5, 2012, Costa Adeje, Spain. New York: IEEE, 2012: 234-239.
[13]	PEVNÝ T, FILLER T, BAS P. Using High-dimensional Image Models to Perform Highly Undetectable Steganography[C]//Information Hiding. International Workshop on Information Hiding, Jun 28-30, 2010, Calgary, Canada. Heidelberg: Springer, 2010: 161-177.
[14]	LI Bin, WANG Ming, HUANG Jiwu, et al. A New Cost Function for Spatial Image Steganography[C]//IEEE. 2014 IEEE International Conference on Image Processing(ICIP), October 27-30, 2014, Paris, France. New York: IEEE, 2014: 4206-4210.
[15]	KODOVSKY J, FRIDRICH J, HOLUB V. Ensemble Classifiers for Steganalysis of Digital Media[J]. IEEE Transactions on Information Forensics and Security, 2011, 7(2):432-444. doi: 10.1109/TIFS.2011.2175919 URL
[16]	ZHANG Ru, ZHU Feng, LIU Jianyi, et al. Depth-wise Separable Convolutions and Multi-level Pooling for An Efficient Spatial CNN-based Steganalysis[EB/OL]. https://ieeexplore.ieee.org/abstract/document/8809687, 2019-08-22.
[17]	WANG Yaojie, NIU Ke, YANG Xiaoyuan. Image Steganography Scheme Based on GANs[J]. Netinfo Security, 2019, 19(5):54-60.
	王耀杰, 钮可, 杨晓元. 基于生成对抗网络的图像隐藏方案[J]. 信息网络安全, 2019, 19(5):54-60.
[18]	MOOSAVI-DEZFOOLI S M, FAWZI A, FAWZI O, et al. Universal Adversarial Perturbations[C]//IEEE. Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition(CVPR), July 21-26, 2017, Honolulu, HI, USA. New York: IEEE, 2014: 4206-4210.
[19]	SU J, VARGAS D V, SAKURAI K. One Pixel Attack for Fooling Deep Neural Networks[J]. IEEE Transactions on Evolutionary Computation, 2019, 23(5):828-841. doi: 10.1109/TEVC.4235 URL
[20]	ZHOU Lingchen, FENG Guorui, SHEN Liquan, et al. On Security Enhancement of Steganography via Generative Adversarial Image[J]. IEEE Signal Processing Letters, 2019, 27(99):166-170. doi: 10.1109/LSP.97 URL
[21]	GOODFELLOW I, POUGET-ABADIE J, MIRZA M, et al. Generative Adversarial Nets[C]// International Conference on Neural Information Processing Systems. Proceedings of the 27th International Conference on Neural Information Processing Systems, December 8-13, 2014, Montreal, Canada. Cambridge: MIT Press, 2014: 2672-2680.

	S-UNIWARD^[3]（未增强）			ZHANG^[10]			ZHOU^[20]			本文的算法
	${{P}_{FA}}$	${{P}_{MD}}$	${{P}_{E}}$	${{P}_{FA}}$	${{P}_{MD}}$	${{P}_{E}}$	${{P}_{FA}}$	${{P}_{MD}}$	${{P}_{E}}$	${{P}_{FA}}$	${{P}_{MD}}$	${{P}_{E}}$
SRNet	0.127	0.137	0.132	0.005	0.999	0.502	0.998	0.001	0.500	0.019	0.983	0.501
YeNet	0.186	0.196	0.191	0.006	0.992	0.499	0.880	0.201	0.540	0.012	0.993	0.503
ZhuNet	0.195	0.214	0.205	0.006	0.999	0.503	0.865	0.128	0.497	0.012	0.984	0.498

	ZHANG^[10]		ZHOU^[20]		本文的算法
	网络训练总时间/s	对抗噪声构建总时间/s	网络训练总时间/s	对抗噪声构建总时间/s	网络训练总时间/s	对抗噪声构建总时间/s
SRNet	-	15,354.07	51,184.59	822.87	-	30.52
YeNet	-	14,968.02	49,162.22	797.19	-	15.91
ZhuNet	-	23,560.27	50,120.15	792.99	-	55.27

	S-UNIWARD^[3] （未增强）	ZHANG^[10]	ZHOU^[20]	本文算法
SRNet	0.132	0.268	0.445	0.497
YeNet	0.188	0.349	0.415	0.498
ZhuNet	0.205	0.257	0.402	0.500

目标隐写分析器用于检测的隐写分析器	SRNet	YeNet	ZhuNet
SRNet	-	0.390	0.425
YeNet	0.293	-	0.471
ZhuNet	0.355	0.467	-

目标隐写分析器用于检测的隐写分析器	SRNet + YeNet	SRNet+ ZhuNet	YeNet+ ZhuNet	SRNet + ZhuNet+ YeNet
SRNet	0.500	0.500	0.499	0.500
YeNet	0.501	0.498	0.500	0.500
ZhuNet	0.507	0.502	0.503	0.501