Generative Steganography Method Based on Diffusion Model and Generative Adversarial Network

doi:10.3969/j.issn.1671-1122.2025.08.004

Abstract

Abstract:

Generative steganography is an emerging technology that encodes secret messages directly into steganographic images, typically built upon existing image generation frameworks such as generative adversarial networks (GAN) and flow models. However, mainstream generative steganography methods currently exhibit significant shortcomings in two critical dimensions: the accuracy of secret information extraction and image quality. In recent years, diffusion models, as a new generation of image generation technology, have provided novel approaches to addressing this technical bottleneck.This paper proposed a generative steganography method that integrated the denoising diffusion implicit model (DDIM) with GANs. Firstly, encoding the secret message into a Gaussian noise space using GANs. Secondly, transforming the noise into steganographic images via DDIM. Finally, leveraging DDIM's determinism, reversibility, and autoencoder structure to efficiently extract the secret message from the image.Experimental results demonstrate that the proposed method outperforms existing solutions across core metrics, including steganographic security, extraction accuracy, and image quality.

Key words: generative steganography, diffusion model, generative adversarial network

CLC Number:

TP309

XIONG Ao, LIU Yuxiao, QIAN Xusheng, ZHANG Nan. Generative Steganography Method Based on Diffusion Model and Generative Adversarial Network[J]. Netinfo Security, 2025, 25(8): 1223-1230.

Figures/Tables 8

References 24

[1]	HU Donghui, WANG Liang, JIANG Wenjie, et al. A Novel Image Steganography Method via Deep Convolutional Generative Adversarial Networks[J]. IEEE Access, 2018, 6: 38303-38314.
[2]	YANG Kuan, CHEN Kejiang, ZHANG Weiming, et al. Provably Secure Generative Steganography Based on Autoregressive Model[C]// Springer. The 17th International Workshop on Digital Forensics and Watermarking. Heidelberg: Springer, 2018: 55-68.
[3]	LIU Xiyao, MA Ziping, MA Junxing, et al. Image Disentanglement Autoencoder for Steganography without Embedding[C]// IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 2303-2312.
[4]	WEI Ping, LI Shen, ZHANG Xinpeng, et al. Generative Steganography Network[C]// ACM. The 30th ACM International Conference on Multimedia. New York: ACM, 2022: 1621-1629.
[5]	ZHOU Zhili, SU Yuecheng, LI Jin, et al. Secret-to-Image Reversible Transformation for Generative Steganography[J]. IEEE Transactions on Dependable and Secure Computing, 2022, 20: 4118-4134.
[6]	SONG Jiaming, MENG Chenlin, ERMON S. Denoising Diffusion Implicit Models[EB/OL]. (2021-01-13)[2024-03-01]. https://openreview.net/forum?id=St1giarCHLP.
[7]	HO J, JAIN A, ABBEEL P. Denoising Diffusion Probabilistic Models[EB/OL]. (2020-12-16)[2024-03-01]. https://arxiv.org/abs/2006.11239.
[8]	MIELIKAINEN J. LSB Matching Revisited[J]. IEEE Signal Processing Letters, 2006, 13(5): 285-287.
[9]	PEVNY T, FILLER T, BAS P. Using High-Dimensional Image Models to Perform Highly Undetectable Steganography[J]. Lecture Notes in Computer Science, 2010, 6387: 161-177.
[10]	HOLUB V, FRIDRICH J. Designing Steganographic Distortion Using Directional Filters[C]// IEEE. 2012 IEEE International Workshop on Information Forensics and Security (WIFS). New York: IEEE, 2012: 234-239.
[11]	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.
[12]	TANG Weixuan, TAN Shunquan, LI Bin, et al. Automatic Steganographic Distortion Learning Using a Generative Adversarial Network[J]. IEEE Signal Processing Letters, 2017, 24(10): 1547-1551.
[13]	BALUJA S. Hiding Images in Plain Sight: Deep Steganography[C]// NIPS. The 31st International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2017: 2066 - 2076.
[14]	GUAN Zhenyu, JING Junpeng, DENG Xin, et al. DeepMIH: Deep Invertible Network for Multiple Image Hiding[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 45(1): 372-390.
[15]	JING Junpeng, DENG Xin, XU Mai, et al. HiNet: Deep Image Hiding by Invertible Network[C]// IEEE. 2021 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE, 2021: 4713-4722.
[16]	LU Shaoping, WANG Rong, ZHONG Tao, et al. Large-Capacity Image Steganography Based on Invertible Neural Networks[C]// IEEE. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2021: 10811-10820.
[17]	XU Youmin, MOU Chong, HU Yujie, et al. Robust Invertible Image Steganography[C]// IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 7865-7874.
[18]	YU Cong, HU Donghui, ZHENG Shuli, et al. An Improved Steganography without Embedding Based on Attention GAN[J]. Peer-to-Peer Networking and Applications, 2021, 14: 1446-1457.
[19]	KRIZHEVSKY A. Learning Multiple Layers of Features from Tiny Images[EB/OL]. (2009-04-08)[2024-03-01]. https://www.cs.toronto.edu/-kriz/learning-features-2009-TR.pdf.
[20]	KARRAS T, LAINE S, AILA T. A Style-Based Generator Architecture for Generative Adversarial Networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(12): 4217-4228.
[21]	CHOI Y, UH Y, YOO J, et al. StarGAN V2: Diverse Image Synthesis for Multiple Domains[C]// IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 8185-8194.
[22]	XU Guanshuo, WU Hanzhou, SHI Yunqing. Structural Design of Convolutional Neural Networks for Steganalysis[J]. IEEE Signal Processing Letters, 2016, 23(5): 708-712.
[23]	HEUSEL M, RAMSAUER H, UNTERTHINER T, et al. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium[C]// NIPS. The 31st International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2017: 6629-6640.
[24]	KINGMA D P, DHARIWAL P. Glow:Generative Flow with Invertible 1x1 Convolutions[C]// NIPS. The 32nd International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2018: 10236-10245.

bpp	1.0			2.0			3.0
指标	Acc	Pe	FID	Acc	Pe	FID	Acc	Pe	FID
GSN	97.15%	0.183	13.40	79.62%	0.022	24.80	72.74%	0.049	30.40
S2IRT	99.94%	0.003	72.60	97.79%	0.002	78.20	97.56%	0.003	67.80
本文方法	98.32%	0.427	2.77	98.54%	0.261	3.30	98.43%	0.310	4.30

bpp	1.0			2.0			3.0
指标	Acc	Pe	FID	Acc	Pe	FID	Acc	Pe	FID
GSN	96.64%	0.174	14.70	78.45%	0.035	23.90	71.36%	0.062	31.80
S2IRT	99.92%	0.006	73.90	96.32%	0.004	75.20	96.13%	0.007	76.60
本文方法	98.71%	0.483	2.54	98.34%	0.262	3.26	98.18%	0.332	4.70

bpp	1.0			2.0			3.0
指标	Acc	Pe	FID	Acc	Pe	FID	Acc	Pe	FID
GSN	96.23%	0.204	13.80	79.02%	0.036	23.70	70.44%	0.056	29.50
S2IRT	99.97%	0.002	73.80	97.48%	0.005	77.60	96.42%	0.008	78.30
本文方法	98.39%	0.446	2.23	98.74%	0.285	3.99	98.78%	0.327	4.23

方法	bpp	1.0	2.0	3.0	4.0	5.0	6.0
AFHQv2 64×64	Acc	98.03%	98.57%	99.19%	96.39%	93.15%	91.93%
	Pe	0.396	0.388	0.376	0.364	0.390	0.394
	FID	2.21	2.35	2.46	2.42	2.35	2.34
FFHQ 64×64	Acc	98.32%	98.54%	98.43%	96.47%	93.28%	92.02%
	Pe	0.427	0.261	0.310	0.334	0.345	0.385
	FID	2.77	3.30	4.30	3.90	3.67	3.37
CIFAR-10	Acc	95.38%	95.07%	95.19%	89.93%	86.43%	84.83%
	Pe	0.434	0.460	0.434	0.417	0.446	0.441
	FID	2.09	2.30	2.66	2.48	2.37	2.31