| 引用本文： |
-
吕美杨,易小伟,符皓程,曹纭,刘长军.基于社交网络无通联模型的鲁棒图像隐写方法[J].信息安全学报,已采用 [点击复制]
- Lv Meiyang,Yi Xiaowei,Fu Haocheng,Cao Yun,Liu Changjun.A Robust Image Steganography Method Based on a Non-communicative Model for Social Networks[J].Journal of Cyber Security,Accept [点击复制]
|
|
| 本文已被：浏览 76次 下载 0次 |
|
| 基于社交网络无通联模型的鲁棒图像隐写方法 |
|
吕美杨1,2, 易小伟1,2, 符皓程1,2, 曹纭1,2, 刘长军1,2
|
|
|
| (1.中国科学院信息工程研究所;2.中国科学院大学网络空间安全学院) |
|
| 摘要: |
| 随着社交网络的迅速发展，利用公开社交信道进行安全隐蔽通信越来越流行，但是当前隐写信息提取依赖于文件分享和下载等异常方式，难以抵御通信关联分析，极大地破坏了通信行为的隐蔽性。针对现有的生成式隐写方法难以抵抗社交信道的复合有损处理，以及隐写网络结构复杂、训练代价高等问题，本文设计了基于社交内容分享的无通联隐写通信模型，并基于此模型提出了一种高安全鲁棒的图像隐写方法，实现经过社交信道传输和拍摄等有损处理后隐藏信息的鲁棒提取，并在保证安全性的同时，降低了隐写网络训练代价。首先，在隐写网络中设计了鲁棒增强层，通过在训练过程中模拟社交信道常见的图像处理操作，增强方法抵抗数字失真的能力，弱化了通信双方的通联关系，增强了通信行为的隐蔽性和安全性。其次，通过模拟屏幕拍摄过程中产生的色彩偏移、图像扭曲等物理失真现象，实现支持屏拍方式的鲁棒提取，使得通信双方无需构建通联关系，进一步增强了通信行为的隐蔽性和抗关联分析能力。最后，通过将映射后的秘密信息嵌入到生成模型编码的隐空间向量，降低了网络结构的复杂度和训练代价。实验结果表明，本文所提方法能达到更优的鲁棒性、图像质量和安全性，经过微博、Facebook等社交信道传输和拍摄处理后，隐藏信息的提取准确率达到97%以上。消融实验证明，本文设计的鲁棒增强层能够有效增强隐写方法的鲁棒性和安全性。 |
| 关键词: 图像隐写 生成模型 社交信道 鲁棒性 无通联 拍摄提取 |
| DOI： |
| 投稿时间：2025-05-14修订日期：2025-11-17 |
| 基金项目:国家自然科学基金项目（面上项目，重点项目，重大项目） |
|
| A Robust Image Steganography Method Based on a Non-communicative Model for Social Networks |
|
Lv Meiyang1,2, Yi Xiaowei1,2, Fu Haocheng1,2, Cao Yun1,2, Liu Changjun1,2
|
| (1.Institute of Information Engineering, Chinese Academy of Sciences;2.School of Cyber Security, University of Chinese Academy of Sciences) |
| Abstract: |
| With the rapid development of social networks, secure covert communication via public social channels has become in-creasingly prevalent. However, current steganographic information extraction relies on anomalous methods such as file sharing and downloading, rendering it vulnerable to communication correlation analysis and significantly compromising the concealment of communication activities. Addressing the limitations of existing generative steganography meth-ods—which struggle against composite lossy processing in social channels and suffer from complex network structures and high training costs—this paper designs a communication model for non-correlated steganography based on social content sharing. Building upon this model, we propose a highly secure and robust image steganography method. This enables the robust extraction of hidden information after lossy processing, such as transmission through social channels or photographic capture, while simultaneously reducing steganography network training costs without compromising security. Firstly, a robust enhancement layer is incorporated into the steganographic network. By simulating common im-age processing operations in social channels during training, this layer bolsters resistance to digital distortion, diminishes the need for prior communication agreements between parties, and enhances the concealment and security of communica-tion activities. Secondly, by simulating physical distortions such as colour shifts and image warping during screen cap-ture, robust extraction is achieved for screen-capture scenarios. This eliminates the need for communication parties to establish a connection, further enhancing the concealment and resistance to association analysis of communication activ-ities. Finally, embedding the mapped secret information into the latent space vectors encoded by the generative model reduces the complexity of the network architecture and the training cost. Experimental results demonstrate that the pro-posed method achieves superior robustness, image quality, and security. After transmission via social channels such as Weibo and Facebook, followed by image processing, the extraction accuracy of hidden information exceeds 97%. Abla-tion studies confirm that the designed robust enhancement layer effectively bolsters the robustness and security of the steganography method. |
| Key words: image steganography generative model social channels robustness Non-communicative photographic extraction |
