Composition-Aware Image Steganography through …

Composition-Aware Image Steganography throughAdversarial Self-Generated Supervision

Zheng Ziqiang 1 Yuanmeng Hu 2 Yi Bin 1 hmk and Yang Yang

1) School of Computer Science and Engineering, University of Electronic Science andTechnology of China

2) Department of Mathematics, Pusan National University

Corresponding Author : Heng Tao Shen

ABSTRACTSteganography is an important and prevailing information hiding tool to perform secret messagetransmission in an open environment. Existing steganography methods can mainly fall into twocategories: pre-defined rule-based and data driven methods. The former is susceptible to thestatistical attack while the latter adopts the deep convolution neural networks to promote secu-rity under statistical attack. However, the deep learning-based methods suffer from perceptibleartificial artifacts. In this paper, we introduce a novel Composition-Aware Image Steganographytermed CAIS to guarantee both visual security and robustness to attack through self-generatedsupervision. The key innovation is an adversarial composition estimation module to integraterule based and deep generative adversarial methods. We perform a rule-based image blend-ing method to obtain infinite synthetically data-label pairs and perform an auxiliary adversarialcomposition estimation task. The innovative self-generated supervision could largely promotethe ability to recognize message patterns from steganographic outputs, which results in bettersteganography performance. Furthermore, an effective Global-and-Part checking is designedto alleviate visual artifacts caused by hiding secret information. We conduct a comprehensiveanalysis of CAIS from various aspects such as security and robustness to verify the superior per-formance of the proposal. Experimental results on three large scale widely-used datasets showthe superior performance of our CAIS compared with several state-of-the-art approaches.

EQUATIONS, TABLES AND FIGURES

Some important framework and experimental results table are listed as following:

Alice BobEve

A beautiful flower?

Open environment

ReconstructedmessageSteganographic

output

Privatemessage

Coverimage

Hidingprocess

Revealingprocess

Intercept

Figure 1. Alice intends to send her portrait to Bob by transmitting it in an open communitywhile she does not want anyone other than Bob access to her photo. So Alice randomly selectsa cover image and hides the message image (her portrait) to generate a steganographic image,which is visually identical to the cover image. Even this steganographic image is intercepted bya third party like Eve, no personal information of Alice will be recognized.

Figure 2. The four components of the full system. Upper Left Corner: Hiding message im-age xm in cover image xc through steganography function F and synthesizing steganographicoutput xc. Lower left corner: Generating composite image xw by Alpha blending and usingself-generated data-label: (xw, α) to optimize the auxiliary estimator of Dg. Lower Right Cor-ner: Randomly cropping part regions from xc and xc and performing part checking throughpart discriminator Dp. Upper Right Corner: Uncovering the steganographic image with therevealing network G to a reconstructed image xm.

REFERENCES

1. Rameen Abdal, Yipeng Qin, and Peter Wonka. Image2stylegan: How to embed images intothe stylegan latent space? In IEEE International Conference on Computer Vision (ICCV),pages 4432–4441, 2019.

2. Anna Babu and Sonal Ayyappan. A reversible crypto-watermarking system for secure

Table 1 Quantitative comparison of Within-domain image steganography between differentmethods. ↑ (↓) indicates that the larger (smaller) the value is, the better the performance.

MethodsSteganography

MSE ↓ RMSE ↓ PSNR ↑ SSIM ↑ S↑ N↑ Q↑ Human↓

Steganography [3] 0.0022 0.0462 26.78 0.9822 0.1997 0.5860 0.6210 94.7

Deep-stegano [49] 0.0015 0.0370 29.01 0.9821 0.2016 0.5917 0.6237 78.9

ISGAN [67] 0.0014 0.0395 29.16 0.9815 0.2031 0.5978 0.6243 71.3

CAIS 0.0011 0.0297 31.07 0.9836 0.2178 0.6287 0.6417 64.3

MethodsReconstruction

MSE ↓ RMSE ↓ PSNR ↑ SSIM ↑

Steganography [3] 0.0022 0.0467 26.66 0.9742

Deep-stegano [49] 0.0026 0.0493 26.37 0.9788

ISGAN [67] 0.0037 0.0596 25.17 0.9742

CAIS 0.0014 0.0347 29.58 0.9861

Steganography Deep-stegano ISGAN CAIS Cover Steganography Deep-stegano ISGAN CAIS Message

Figure 3. The visual within-domain image steganography comparison of different methods. Theimages at the left side of black dotted show the steganography comparison while the images atthe right side of black dotted show the reconstruction comparison.

medical image transmission. In Annual IEEE India Conference (INDICON), pages 1–6.IEEE, 2015.

3. Shumeet Baluja. Hiding images in plain sight: Deep steganography. In Advances in NeuralInformation Processing Systems (NeurIPS), pages 2069–2079, 2017.

4. Long Bao and Yicong Zhou. Image encryption: Generating visually meaningful encryptedimages. Information Sciences, 324:197–207, 2015.

5. Mihir Bellare and Phillip Rogaway. Optimal asymmetric encryption. In Workshop on theTheory and Application of of Cryptographic Techniques, pages 92–111. Springer, 1994.

6. Robert AA Campbell, Robert W Eifert, and Glenn C Turner. Openstage: a low-costmotorized microscope stage with sub-micron positioning accuracy. PloS one, 9(2), 2014.

7. Mathilde Caron, Piotr Bojanowski, Armand Joulin, and Matthijs Douze. Deep clustering

Table 2 Quantitative comparison of Cross-domain image steganography between differentmethods. ↑ (↓) indicates that the larger (smaller) the value is, the better the performance.

MethodsSteganography

MSE ↓ RMSE ↓ PSNR ↑ SSIM ↑ S↑ N↑ Q↑ Human↓

Steganography [3] 0.0027 0.0512 25.90 0.9876 0.2034 0.5790 0.6227 98.4

Deep-stegano [49] 0.0025 0.0497 27.76 0.9856 0.2087 0.5842 0.6276 83.1

ISGAN [67] 0.0014 0.0384 29.18 0.9814 0.2093 0.5813 0.6218 76.3

CAIS 0.0010 0.0151 30.73 0.9845 0.2172 0.5829 0.6331 61.6

MethodsReconstruction

MSE ↓ RMSE ↓ PSNR ↑ SSIM ↑

Steganography [3] 0.0032 0.0565 25.00 0.9755

Deep-stegano [49] 0.0029 0.0521 26.65 0.9765

ISGAN [67] 0.0029 0.0542 26.14 0.9713

CAIS 0.0023 0.0439 27.70 0.9853

Steganography Deep-stegano ISGAN CAIS Cover Steganography Deep-stegano ISGAN CAIS Message

Figure 4. The visual cross-domain image steganography results of different methods. The im-ages at the left side of black dotted show the steganography comparison while the images at theright side of black dotted show the reconstruction comparison.

for unsupervised learning of visual features. In IEEE European Conference on ComputerVision (ECCV), pages 132–149, 2018.

8. NK Cauvery. Water marking on digital image using genetic algorithm. InternationalJournal of Computer Science Issues (IJCSI), 8(6):323, 2011.

9. Abbas Cheddad, Joan Condell, Kevin Curran, and Paul Mc Kevitt. Digital imagesteganography: Survey and analysis of current methods. Signal processing, 90(3):727–752,2010.

10. Kejiang Chen, Hang Zhou, Wenbo Zhou, Weiming Zhang, and Nenghai Yu. Definingcost functions for adaptive jpeg steganography at the microscale. IEEE Transactions onInformation Forensics and Security (TIFS), 14(4):1052–1066, 2018.

11. Qifeng Chen and Vladlen Koltun. Photographic image synthesis with cascaded refinementnetworks. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),pages 1511–1520, 2017.

12. Yunjey Choi, Minje Choi, Munyoung Kim, Jung-Woo Ha, Sunghun Kim, and JaegulChoo. StarGAN: Unified generative adversarial networks for multi-domain image-to-image translation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), pages 8789–8797, 2018.

13. Casey Chu, Andrey Zhmoginov, and Mark Sandler. CycleGAN, a master of steganography.arXiv preprint arXiv:1712.02950, 2017.

14. Daniel-L. Currie and Cynthia-E. Irvine. Surmounting the effects of lossy compression onsteganography, 1996.

15. Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: Alarge-scale hierarchical image database. In IEEE/CVF Conference on Computer Visionand Pattern Recognition (CVPR), pages 248–255. Ieee, 2009.

16. Alexey Dosovitskiy and Thomas Brox. Generating images with perceptual similaritymetrics based on deep networks. In Advances in Neural Information Processing Systems(NeurIPS), pages 658–666, 2016.

17. Khalid El Makkaoui, Abdellah Ezzati, and Abderrahim Beni-Hssane. Cloud-rsa: anenhanced homomorphic encryption scheme. In Europe and MENA Cooperation Advancesin Information and Communication Technologies, pages 471–480. Springer, 2017.

18. Amin Milani Fard, M-R Akbarzadeh-T, and Farshad Varasteh-A. A new genetic algorithmapproach for secure jpeg steganography. In IEEE International Conference on IntelligentSystems Engineering (ICIES), pages 1–6. IEEE, 2006.

19. Prabakaran Ganesan and R Bhavani. A high secure and robust image steganography usingdual wavelet and blending model. JCS, 9(3):277–284, 2013.

20. Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, SherjilOzair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances inNeural Information Processing Systems (NeurIPS), pages 2672–2680, 2014.

21. Chen Guanrong, Mao Yaobin, and Charles K Chui. A symmetric image encryption schemebased on 3d chaotic cat maps. Chaos, Solitons and Fractals, 21(3):749–761, 2004.

22. Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrastfor unsupervised visual representation learning. In IEEE/CVF Conference on ComputerVision and Pattern Recognition (CVPR), pages 9729–9738, 2020.

23. Martin Heusel, Hubert Ramsauer, Thomas Unterthiner, Bernhard Nessler, and SeppHochreiter. GANs trained by a two time-scale update rule converge to a local Nashequilibrium. In Advances in Neural Information Processing Systems (NeurIPS), pages6626–6637, 2017.

24. Sabah Husien and Haitham Badi. Artificial neural network for steganography. NeuralComputing and Applications, 26(1):111–116, 2015.

25. Mehdi Hussain, Ainuddin Wahid Abdul Wahab, Yamani Idna Bin Idris, Anthony TS Ho,and Ki-Hyun Jung. Image steganography in spatial domain: A survey. Signal Processing:Image Communication, 65:46–66, 2018.

26. Phillip Isola, Jun-Yan Zhu, Tinghui Zhou, and Alexei A Efros. Image-to-image translationwith conditional adversarial networks. In IEEE/CVF Conference on Computer Vision andPattern Recognition (CVPR), pages 1125–1134, 2017.

27. A Joseph and V Sundaram. Cryptography and steganography–a survey. 2011.

28. Kamaldeep Joshi and Rajkumar Yadav. A new lsb-s image steganography method blendwith cryptography for secret communication. In International Conference on ImageInformation Processing (ICIIP), pages 86–90. IEEE, 2015.

29. Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen. Progressive growing of gansfor improved quality, stability, and variation. arXiv preprint arXiv:1710.10196, 2017.

30. Yan Ke, Minqing Zhang, Jia Liu, Tingting Su, and Xiaoyuan Yang. Generativesteganography with kerckhoffs’ principle based on generative adversarial networks. arXivpreprint arXiv:1711.04916, 2017.

31. Gary C Kessler and Chet Hosmer. An overview of steganography. In Advances inComputers, volume 83, pages 51–107. Elsevier, 2011.

32. Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXivpreprint arXiv:1412.6980, 2014.

33. Erik Learned-Miller, Gary B Huang, Aruni RoyChowdhury, Haoxiang Li, and Gang Hua.Labeled faces in the wild: A survey. In Advances in Face Detection and Facial ImageAnalysis, pages 189–248. Springer, 2016.

34. Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham,Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, et al.Photo-realistic single image super-resolution using a generative adversarial network. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 4681–4690, 2017.

35. Bin Li, Junhui He, Jiwu Huang, and Yun Qing Shi. A survey on image steganography andsteganalysis. Information Hiding and Multimedia Signal Processing, 2(2):142–172, 2011.

36. Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes inthe wild. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR),pages 3730–3738, 2015.

37. Xiangyang Luo, Fenlin Liu, Chunfang Yang, Shiguo Lian, and Daoshun Wang. On f5steganography in images. The Computer Journal, 55(4):447–456, 2012.

38. Alexander Mansurov. A ctf-based approach in information security education: anextracurricular activity in teaching students at altai state university, russia. Modern AppliedScience, 10(11):159, 2016.

39. Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. arXiv preprintarXiv:1411.1784, 2014.

40. Tayana Morkel, Jan HP Eloff, and Martin S Olivier. An overview of image steganography.In International Social Security Association (ISSA), volume 1, 2005.

41. Maria-Elena Nilsback and Andrew Zisserman. Automated flower classification over alarge number of classes. In ndian Conference on Computer Vision, Graphics and ImageProcessing (ICVGIP), pages 722–729. IEEE, 2008.

42. Lionel Pibre, Pasquet Jerome, Dino Ienco, and Marc Chaumont. Deep learning forsteganalysis is better than a rich model with an ensemble classifier, and is natively robustto the cover source-mismatch. arXiv preprint arXiv:1511.04855, 2015.

43. Lionel Pibre, Jerome Pasquet, Dino Ienco, and Marc Chaumont. Deep learning is a goodsteganalysis tool when embedding key is reused for different images, even if there is a coversourcemismatch. Electronic Imaging, 2016(8):1–11, 2016.

44. Vidyasagar M Potdar, Song Han, and Elizabeth Chang. A survey of digital imagewatermarking techniques. In IEEE International Conference on Industrial Informatics(INDIN), pages 709–716. IEEE, 2005.

45. Niels Provos and Peter Honeyman. Detecting steganographic content on the internet.Technical report, Center for Information Technology Integration, 2001.

46. Yinlong Qian, Jing Dong, Wei Wang, and Tieniu Tan. Deep learning for steganalysis viaconvolutional neural networks. In Media Watermarking, Security, and Forensics 2015,volume 9409, page 94090J. International Society for Optics and Photonics, 2015.

47. Yuhui Quan, Huan Teng, Yixin Chen, and Hui Ji. Watermarking deep neural networks inimage processing. IEEE Transactions on Neural Networks and Learning Systems (TNNLS),2020.

48. Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learningwith deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434,2015.

49. Rafia Rahim, Shahroz Nadeem, et al. End-to-end trained cnn encoder-decoder networksfor image steganography. In IEEE European Conference on Computer Vision (ECCV)Workshops (ECCVW), pages 0–0, 2018.

50. KB Raja, CR Chowdary, KR Venugopal, and LM Patnaik. A secure image steganographyusing lsb, dct and compression techniques on raw images. In International Conference onIntelligent Sensing and Information Processing (ICISIP), pages 170–176, 2005.

51. U Reethika and Mrs S Anitha. An overview of steganography and cryptography.

52. Yang Ren-Er, Zheng Zhiwei, Tao Shun, and Ding Shilei. Image steganography combinedwith des encryption pre-processing. In International Conference on Measuring Technologyand Mechatronics Automation (ICMTMA), pages 323–326. IEEE, 2014.

53. O Ronneberger, P Fischer, and TU-net Brox. Convolutional networks for biomedical imagesegmentation. In International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI).

54. Kuniaki Saito, Yoshitaka Ushiku, Tatsuya Harada, and Kate Saenko. Strong-weakdistribution alignment for adaptive object detection. In IEEE/CVF Conference on ComputerVision and Pattern Recognition (CVPR), pages 6956–6965, 2019.

55. William Stallings. Cryptography and Network Security: Principles and Practice. PrenticeHall Press, USA, 6th edition, 2013.

56. Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna.Rethinking the inception architecture for computer vision. In IEEE/CVF Conference onComputer Vision and Pattern Recognition (CVPR), pages 2818–2826, 2016.

57. Yi-Hsuan Tsai, Wei-Chih Hung, Samuel Schulter, Kihyuk Sohn, Ming-Hsuan Yang,and Manmohan Chandraker. Learning to adapt structured output space for semanticsegmentation. In IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), pages 7472–7481, 2018.

58. Denis Volkhonskiy, Boris Borisenko, and Evgeny Burnaev. Generative adversarialnetworks for image steganography. 2016.

59. Ekta Walia, Payal Jain, and Navdeep Navdeep. An analysis of lsb & dct basedsteganography. Global Journal of Computer Science and Technology, 2010.

60. Shiping Wen, Zhigang Zeng, Tingwen Huang, Qinggang Meng, and Wei Yao. Lagsynchronization of switched neural networks via neural activation function and applicationsin image encryption. IEEE Transactions on Neural Networks and Learning Systems(TNNLS), 26(7):1493–1502, 2015.

61. Pooja Yadav, Nishchol Mishra, and Sanjeev Sharma. A secure video steganography withencryption based on lsb technique. In IEEE International Conference on ComputationalIntelligence and Computing Research (ICCIC), pages 1–5. IEEE, 2013.

62. Jianhua Yang, Kai Liu, Xiangui Kang, Edward K Wong, and Yun-Qing Shi. Spatial imagesteganography based on generative adversarial network. arXiv preprint arXiv:1804.07939,2018.

63. Hojatollah Yeganeh and Zhou Wang. Objective quality assessment of tone-mapped images.IEEE Transactions on Image Processing (TIP), 22(2):657–667, 2012.

64. Massimiliano Zanin and Alexander N Pisarchik. Gray code permutation algorithm forhigh-dimensional data encryption. Information Sciences, 270:288–297, 2014.

65. Kevin Alex Zhang, Alfredo Cuesta-Infante, Lei Xu, and Kalyan Veeramachaneni.Steganogan: High capacity image steganography with gans. arXiv preprintarXiv:1901.03892, 2019.

66. Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. Theunreasonable effectiveness of deep features as a perceptual metric. In IEEE/CVFConference on Computer Vision and Pattern Recognition (CVPR), pages 586–595, 2018.

67. Ru Zhang, Shiqi Dong, and Jianyi Liu. Invisible steganography via generative adversarialnetworks. Multimedia Tools and Applications, 78(7):8559–8575, 2019.

68. Ying-Qian Zhang and Xing-Yuan Wang. A symmetric image encryption algorithm basedon mixed linear–nonlinear coupled map lattice. Information Sciences, 273:329–351, 2014.

69. Sicheng Zhao, Xiangyu Yue, Shanghang Zhang, Bo Li, Han Zhao, Bichen Wu, RaviKrishna, Joseph E Gonzalez, Alberto L Sangiovanni-Vincentelli, Sanjit A Seshia, et al.A review of single-source deep unsupervised visual domain adaptation. IEEE Transactionson Neural Networks and Learning Systems (TNNLS), 2020.

70. Ziqiang Zheng, Hongzhi Liu, Zhibin Yu, Haiyong Zheng, Yang Wu, Yang Yang, andJianbo Shi. Encryptgan: Image steganography with domain transform. arXiv preprintarXiv:1905.11582, 2019.

71. Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A Efros. Unpaired image-to-imagetranslation using cycle-consistent adversarial networks. In IEEE International Conferenceon Computer Vision (ICCV), pages 2223–2232, 2017.

72. Yang Zou, Zhiding Yu, Xiaofeng Liu, BVK Kumar, and Jinsong Wang. Confidenceregularized self-training. In IEEE International Conference on Computer Vision (ICCV),pages 5982–5991, 2019.