基于扩散模型和生成对抗网络的生成式隐写方法

doi:10.3969/j.issn.1671-1122.2025.08.004

信息网络安全 ›› 2025, Vol. 25 ›› Issue (8): 1223-1230.doi: 10.3969/j.issn.1671-1122.2025.08.004

基于扩散模型和生成对抗网络的生成式隐写方法

熊翱¹, 刘雨潇¹(), 钱旭盛², 张楠³

1.北京邮电大学网络与交换技术全国重点实验室，北京 100876
2.国网江苏省电力有限公司营销服务中心，南京 210024
3.国家电网有限公司客户服务中心，天津 300304

收稿日期:2024-05-27 出版日期:2025-08-10 发布日期:2025-09-09
通讯作者: 刘雨潇 E-mail:liuyuxiao@bupt.edu.cn
作者简介:熊翱（1974—），男，江西，副教授，博士，主要研究方向为网络管理和通信软件|刘雨潇（2000—），男，重庆，硕士研究生，主要研究方向为信息隐藏技术|钱旭盛（1975—），男，江苏，高级工程师，本科，主要研究方向为电力市场理论与策略、电力营销客户服务管理和电子服务渠道运营|张楠（1996—），女，江苏，硕士，主要研究方向为电力营销及大数据
基金资助:
国家重点研发计划(2022YFB2703400);国家电网科技项目(5700-202353318A-1-1-ZN)

Generative Steganography Method Based on Diffusion Model and Generative Adversarial Network

XIONG Ao¹, LIU Yuxiao¹(), QIAN Xusheng², ZHANG Nan³

1. State Key Laboratory of Networking and Switching Technology, Beijing University of Posts and Telecommunications, Beijing 100876, China
2. Marketing Service Center, State Grid Jiangsu Electric Power Co., Ltd., Nanjing 210024, China
3. Customer Service Center, State Grid Corporation of China, Tianjin 300304, China

Received:2024-05-27 Online:2025-08-10 Published:2025-09-09

摘要/Abstract

摘要：

生成式隐写术是一种新兴技术，其核心在于将秘密消息直接编码为隐写图像。该技术通常以现有图像生成模型为基础框架，如生成对抗网络（GAN）和流模型。然而，当前主流的生成式隐写术在秘密信息提取准确率与图像质量两个关键维度上均存在不足。近年来，扩散模型作为新一代图像生成技术，为解决这一技术瓶颈提供了新的思路。文章提出一种融合去噪扩散隐式模型（DDIM）与GAN的生成式隐写方法：首先，通过GAN将秘密消息编码至高斯噪声空间；然后，利用DDIM将噪声转换为隐写图像；最后，依托DDIM的确定性、可逆性及其自编码器结构，高效提取图像中的秘密消息。实验结果表明，该方法在隐写过程的安全性、信息提取准确率以及图像质量等核心指标上均超越现有方案。

关键词: 生成式隐写术, 扩散模型, 生成对抗网络

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

中图分类号:

TP309

熊翱, 刘雨潇, 钱旭盛, 张楠. 基于扩散模型和生成对抗网络的生成式隐写方法[J]. 信息网络安全, 2025, 25(8): 1223-1230.

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.

图/表 8

图1

图2

表1

表2

表3

图3

表4

图4

参考文献 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

基于扩散模型和生成对抗网络的生成式隐写方法

Generative Steganography Method Based on Diffusion Model and Generative Adversarial Network

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 24

相关文章 8

编辑推荐

Metrics

本文评价

[1]	李元诚, 孙鹤洋, 张桐, 张贺方, 杨立群. 基于DACDiff的分布式电源调度控制系统FDIAs防御方法[J]. 信息网络安全, 2025, 25(4): 578-586.
[2]	冯光升, 蒋舜鹏, 胡先浪, 马明宇. 面向物联网的入侵检测技术研究新进展[J]. 信息网络安全, 2024, 24(2): 167-178.
[3]	张国敏, 屠智鑫, 邢长友, 王梓澎, 张俊峰. 基于对抗样本的流量时序特征混淆方法[J]. 信息网络安全, 2024, 24(12): 1882-1895.
[4]	李季瑀, 付章杰, 张玉斌. 一种基于跨域对抗适应的图像信息隐藏算法[J]. 信息网络安全, 2023, 23(1): 93-102.
[5]	郭森森, 王同力, 慕德俊. 基于生成对抗网络与自编码器的网络流量异常检测模型[J]. 信息网络安全, 2022, 22(12): 7-15.
[6]	王耀杰, 杨晓元, 刘文超. 基于数字化卡登格的生成图像隐写方案[J]. 信息网络安全, 2021, 21(2): 70-77.
[7]	雷雨, 刘佳, 李军, 柯彦. 一种基于条件生成对抗网络的图像隐写方法研究与实现[J]. 信息网络安全, 2021, 21(11): 48-57.
[8]	彭中联, 万巍, 荆涛, 魏金侠. 基于改进CGANs的入侵检测方法研究[J]. 信息网络安全, 2020, 20(5): 47-56.