用于失配隐写分析的对抗子领域自适应网络

doi:10.11959/j.issn.2096-109x.2022028

Abstract

Abstract:

Once data in the training and test sets come from different cover sources, that is, under the condition of cover source mismatch, it usually makes the detection accuracy rate of an outstanding steganalysis model to be reduced.In practical applications, the analyzers need to process images collected from the Internet.However, compared with the training set data, these suspicious images are likely to have completely different capture and processing histories, which may lead to the degradation of steganalysis model.It is also why steganalysis tools are difficult to deploy successfully in the real-world applications.To improve the practical application value of steganalysis methods based on deep learning, test sample information is utilized and domain adaptation method is used to solve the problem of cover source mismatch.Regarding the training set data as the source domain and test set data as the target domain, the detection performance of steganalysis models in the target domain is enhanced by minimizing the discrepancy between the feature distribution of source domain and target domain.ASAN (adversarial subdomain adaptation network) was proposed from the perspective of feature generation on the one hand.The source domain features and target domain features generated by the steganalysis model were required to be as similar as possible, so that the discriminator cannot distinguish which domain the features came from.On the other hand, to reduce the difference of feature distribution between domains, the subdomain adaptation method was adopted to reduce the unexpected change of the distribution of related subdomains.The distance between the cover and stego samples was enlarged effectively to improve the classification accuracy.After testing three steganography algorithms on multiple datasets, it is confirmed that the proposed method can effectively improve the detection accuracy rate of the model in the case of dataset mismatch and algorithm mismatch and it can also reduce the negative impact of the mismatch problem of the model.

Key words: image steganalysis, cover source mismatch, adversarial learning, domain adaptation

CLC Number:

TP309

Lei ZHANG, Hongxia WANG. Adversarial subdomain adaptation network for mismatched steganalysis[J]. Chinese Journal of Network and Information Security, 2022, 8(3): 76-86.

Figures/Tables 10

数据集	隐写算法	SRM	SRNet	J-Net	ASAN
UCID v.2	SUNI	56.9%	81.6%	83.0%	$91 . 5 %$
	WOW	55.9%	81.0%	82.3%	$91 . 9 %$
	HUGO	59.2%	78.8%	81.7%	$88 . 7 %$
DIV2K	SUNI	56.0%	83.5%	84.5%	$93 . 4 %$
	WOW	55.0%	80.5%	81.4%	$92 . 0 %$
	HUGO	57.0%	80.1%	81.7%	$90 . 7 %$
Flickr 25K	SUNI	50.5%	87.4%	87.9%	$91 . 6 %$
	WOW	51.5%	84.3%	86.5%	$88 . 1 %$
	HUGO	50.6%	88.4%	91.0%	$95 . 0 %$
平均值		54.7%	82.8%	84.4%	$91 . 4 %$

数据集	隐写算法	SRM	SRNet	J-Net	ASAN
UCIDv.2	SUNI	51.5%	70.3%	74.8%	$80 . 4 %$
	WOW	52.1%	68.2%	67.8%	$84 . 0 %$
	HUGO	53.1%	70.6%	70.7%	$81 . 5 %$
DIV2K	SUNI	52.0%	71.6%	75.9%	$83 . 1 %$
	WOW	51.8%	69.9%	70.4%	$84 . 5 %$
	HUGO	53.0%	70.2%	70.9%	$81 . 1 %$
Flickr 25K	SUNI	50.0%	76.8%	79.6%	$89 . 2 %$
	WOW	50.5%	71.0%	74.4%	$88 . 2 %$
	HUGO	50.0%	78.3%	81.0%	$90 . 5 %$
平均值		51.6%	71.9%	73.9%	$84 . 7 %$

训练集	测试集	SRM	SRNet	J-Net	ASAN
SUNI	WOW	67.4%	73.1%	80.0%	$82 . 2 %$
	HUGO	69.3%	74.6%	78.6%	$79 . 9 %$
WOW	SUNI	71.9%	87.8%	88.2%	$92 . 5 %$
	HUGO	67.7%	84.0%	84.9%	$89 . 1 %$
HUGO	SUNI	67.6%	85.2%	85.3%	$94 . 3 %$
	WOW	62.9%	82.2%	83.1%	$92 . 4 %$
平均值		67.8%	81.2%	83.4%	$88 . 4 %$

训练集	测试集	SRM	SRNet	J-Net	ASAN
SUNI	WOW	57.0%	67.3%	68.7%	$72 . 1 %$
	HUGO	57.9%	68.9%	69.3%	$73 . 4 %$
WOW	SUNI	59.8%	74.9%	74.7%	$85 . 9 %$
	HUGO	56.8%	72.1%	72.2%	$83 . 3 %$
HUGO	SUNI	57.6%	73.2%	74.8%	$75 . 7 %$
	WOW	54.5%	70.7%	71.4%	$77 . 4 %$
平均值		57.3%	71.2%	71.9%	$78 . 0 %$

数据集	隐写算法	ADV	DSAN	ASAN
UCID v.2	SUNI	85.5%	85.1%	$91 . 5 %$
	WOW	87.7%	86.0%	$91 . 9 %$
	HUGO	84.4%	84.6%	$88 . 7 %$
DIV2K	SUNI	86.7%	86.7%	$93 . 4 %$
	WOW	85.4%	85.6%	$92 . 0 %$
	HUGO	85.5%	84.9%	$90 . 7 %$
Flickr 25K	SUNI	90.1%	$92 . 7 %$	91.6%
	WOW	87.3%	$89 . 9 %$	88.1%
	HUGO	93.2%	93.8%	$95 . 0 %$
	平均值	87.3%	87.7%	$91 . 4 %$

数据集	隐写算法	ADV	DSAN	ASAN
UCID v.2	SUNI	75.0%	76.9%	$80 . 4 %$
	WOW	72.7%	77.3%	$84 . 0 %$
	HUGO	72.9%	76.4%	$81 . 5 %$
DIV2K	SUNI	76.1%	80.7%	$83 . 1 %$
	WOW	73.1%	79.5%	$84 . 5 %$
	HUGO	75.0%	77.0%	$81 . 1 %$
Flickr 25K	SUNI	84.7%	84.9%	$89 . 2 %$
	WOW	75.3%	78.5%	$88 . 2 %$
	HUGO	86.4%	85.4%	$90 . 5 %$
	平均值	76.8%	79.6%	$84 . 7 %$

References 39

[1]	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.
[2]	HOLUB V , FRIDRICH J . Designing steganographic distortion using directional filters[C]// Proceedings of 2012 IEEE International Workshop on Information Forensics and Security (WIFS). 2012: 234-239.
[3]	PEVNY T , FILLER T , BAS P . Using high-dimensional image models to perform highly undetectable steganography[C]// Proceedings of International Workshop on Information Hiding.Springer. 2010: 161-177.
[4]	FRIDRICH J , KODOVSKY J . Rich models for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2012,7(3): 868-882.
[5]	SONG X , LIU F , YANG C ,et al. Steganalysis of adaptive JPEG steganography using 2D Gabor filters[C]// Proceedings of the 3rd ACM Workshop on Information Hiding and Multimedia Security. 2015: 15-23.
[6]	FENG G , ZHANG X , REN Y ,et al. Diversity-based cascade filters for JPEG steganalysis[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2019,30(2): 376-386.
[7]	YE J , NI J , YI Y . Deep learning hierarchical representations for image steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2017,12(11): 2545-2557.
[8]	YEDROUDJ M , COMBY F , CHAUMONT M . Yedroudj-net:an efficient CNN for spatial steganalysis[C]// 2018 IEEE International Conference on Acoustics,Speech and Signal Processing (ICASSP). 2018: 2092-2096.
[9]	BOROUMAND M , CHEN M , FRIDRICH J . Deep residual network for steganalysis of digital images[J]. IEEE Transactions on Information Forensics and Security, 2019,14(5): 1181-1193.
[10]	ZHANG R , ZHU F , LIU J ,et al. Depth-wise separable convolutions and multi-level pooling for an efficient spatial CNN-based steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2020,15: 1138-1150.
[11]	YOU W , ZHANG H , ZHAO X . A siamese CNN for image steganalysis[J]. IEEE Transactions on Information Forensics and Security, 2021,16: 291-306.
[12]	沈军, 廖鑫, 秦拯 ,等. 基于卷积神经网络的低嵌入率空域隐写分析[J]. 软件学报, 2021,32(9): 2901-2915.
	SHEN J , LIAO X , QIN Z ,et al. Spatial steganalysis of low embedding rate based on convolutional neural network[J]. Journal of Software, 2021,32(9): 2901-2915.
[13]	JIA J , ZHAI L , REN W ,et al. Transferable heterogeneous feature subspace learning for JPEG mismatched steganalysis[J]. Pattern Recognition, 2020,100: 107105.
[14]	XUE Y , YANG L , WEN J ,et al. A subspace learning-based method for JPEG mismatched steganalysis[J]. Multimedia Tools and Applications, 2019,78(7): 8151-8166.
[15]	YANG Y , KONG X , FENG C . Double-compressed JPEG images steganalysis with transferring feature[J]. Multimedia Tools and Applications, 2018,77(14): 17993-18005.
[16]	FENG C , KONG X W , LI M ,et al. Contribution-based feature transfer for JPEG mismatched steganalysis[C]// Proceedings of 2017 IEEE International Conference on Image Processing (ICIP). 2017: 500-504.
[17]	YANG Y , KONG X , WANG B ,et al. Steganalysis on Internet images via domain adaptive classifier[J]. Neurocomputing, 2019,351: 205-216.
[18]	LONG M , WANG J , DING G ,et al. Adaptation regularization:a general framework for transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2013,26(5): 1076-1089.
[19]	ZHANG X , KONG X , WANG P ,et al. Cover-source mismatch in deep spatial steganalysis[C]// Proceedings of 18th Workshop on Digital Watermarking. 2019: 71-83.
[20]	HU D , MA Z , FAN Y ,et al. Study on the interaction between the cover source mismatch and texture complexity in steganalysis[J]. Multimedia Tools and Applications, 2019,78(6): 7643-7666.
[21]	张逸为, 张卫明, 俞能海 . 针对特定测试样本的隐写分析方法[J]. 软件学报, 2018,29(4): 987-1001.
	ZHANG Y W , ZHANG W M , YU N H . Specific testing sample steganalysis[J]. Journal of Software, 2018,29(4): 987-1001.
[22]	OZCAN S , MUSTACOGLU A F . Transfer learning effects on image steganalysis with pre-trained deep residual neural network model[C]// Proceedings of 2018 IEEE International Conference on Big Data (Big Data). 2018: 2280-2287.
[23]	EL BEJI R , SAIDI M , HERMASSI H ,et al. An improved CNN steganalysis architecture based on “Catalyst Kernels” and transfer learning[C]// Proceedings of International Conference on Digital Economy. 2018: 119-128.
[24]	PEVNY T , FRIDRICH J . Merging Markov and DCT features for multi-class JPEG steganalysis[C]// Proceedings of Security,Steganography,and Watermarking of Multimedia Contents IX. 2007:650503.
[25]	HOLUB V , FRIDRICH J . Low-complexity features for JPEG steganalysis using undecimated DCT[J]. IEEE Transactions on Information Forensics and Security, 2015,10(2): 219-228.
[26]	KODOVSKY J , FRIDRICH J , . Calibration revisited[C]// Proceedings of the 11th ACM Workshop on Multimedia and Security. 2009: 63-74.
[27]	PAN S J , YANG Q . A survey on transfer learning[J]. IEEE Transactions on Knowledge and Data Engineering, 2009,22(10): 1345-1359.
[28]	BORGWARDT K M , GRETTON A , RASCH M J ,et al. Integrating structured biological data by kernel maximum mean discrepancy[J]. Bioinformatics, 2006,22(14): 49-57.
[29]	TZENG E , HOFFMAN J , ZHANG N ,et al. Deep domain confusion:Maximizing for domain invariance[J]. arXiv preprint arXiv:1412.3474, 2014.
[30]	LONG M , CAO Y , WANG J ,et al. Learning transferable features with deep adaptation networks[C]// Proceedings of International Conference on Machine Learning. 2015: 97-105.
[31]	LONG M , ZHU H , WANG J ,et al. Deep transfer learning with joint adaptation networks[C]// Proceedings of International Conference on Machine Learning. 2017: 2208-2217.
[32]	ZHU Y , ZHUANG F , WANG J ,et al. Deep subdomain adaptation network for image classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020,32(4): 1713-1722.
[33]	GANIN Y , USTINOVA E , AJAKAN H ,et al. Domain-adversarial training of neural networks[J]. The Journal of Machine Learning Research, 2016,17(1): 2096-2030.
[34]	YU C , WANG J , CHEN Y ,et al. Transfer learning with dynamic adversarial adaptation network[C]// Proceedings of 2019 IEEE International Conference on Data Mining (ICDM). 2019: 778-786.
[35]	BAS P , FILLER T , PEVNY T , . “Break our steganographic system”:the ins and outs of organizing BOSS[C]// Proceedings of International Workshop on Information Hiding. 2011: 59-70.
[36]	SCHAEFER G , STICH M . UCID:an uncompressed color image database[C]// Proceedings of Storage and Retrieval Methods and Applications for Multimedia. 2004: 472-480.
[37]	AGUSTSSON E , TIMOFTE R . Ntire 2017 challenge on single image super-resolution:Dataset and study[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. 2017: 126-135.
[38]	HUISKES M J , LEW M S . The MIR flickr retrieval evaluation[C]// Proceedings of the 1st ACM International Conference on Multimedia Information Retrieval. 2008: 39-43.
[39]	VAN DER MAATEN L , HINTON G . Visualizing data using t-SNE[J]. Journal of Machine Learning Research, 2008,9(11): 2579-2605.

Metrics

Recommended 0

No Suggested Reading articles found!

网络层	结构	输出尺寸
输入		1×512
第1层	Linear+BN1d+ReLU	1×512
第2层	Linear+BN1d+ReLU	1×512
第3层	Linear+Sigmoid	1

Adversarial subdomain adaptation network for mismatched steganalysis

RichHTML

PDF下载

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 10

References 39

Related Articles 1

Metrics

Recommended 0