计算机应用 ›› 2018, Vol. 38 ›› Issue (10): 2923-2928.DOI: 10.11772/j.issn.1001-9081.2018030666

• 网络空间安全 • 上一篇    下一篇

基于生成对抗网络的信息隐藏方案

王耀杰1,2, 钮可1,2, 杨晓元1,2   

  1. 1. 武警工程大学 密码工程学院, 西安 710086;
    2. 网络与信息安全武警部队重点实验室(武警工程大学), 西安 710086
  • 收稿日期:2018-04-02 修回日期:2018-05-23 出版日期:2018-10-10 发布日期:2018-10-13
  • 通讯作者: 王耀杰
  • 作者简介:王耀杰(1990-),男,河南洛阳人,硕士研究生,主要研究方向:信息安全、深度学习;钮可(1981-),男,浙江湖州人,副教授,博士研究生,主要研究方向:信息安全、信息隐藏;杨晓元(1959-),男,湖南湘潭人,教授,博士生导师,主要研究方向:信息隐藏、密码学。
  • 基金资助:
    国家重点研发计划项目(2017YFB0802000)。

Information hiding scheme based on generative adversarial network

WANG Yaojie1,2, NIU Ke1,2, YANG Xiaoyuan1,2   

  1. 1. College of Cryptographic Engineering, Engineering College of Armed Police Force, Xi'an Shaanxi 710086, China;
    2. Key Laboratory of Network and Information Security Under the Armed Police Force(Engineering College of Armed Police Force), Xi'an Shaanxi 710086, China
  • Received:2018-04-02 Revised:2018-05-23 Online:2018-10-10 Published:2018-10-13
  • Supported by:
    This work is partially supported by the National Key R&D Program of China (2017YFB0802000).

摘要: 针对信息隐藏中含密载体会留有修改痕迹,从根本上难以抵抗基于统计的隐写分析算法检测的问题,提出一种基于生成对抗网络(GAN)的信息隐藏方案。该方案首先利用生成对抗网络中的生成模型G以噪声为驱动生成原始载体信息;其次,使用±1嵌入算法,将秘密消息嵌入到生成的载体信息中生成含密信息;最终,将含密信息与真实图像样本作为生成对抗网络中判别模型D的输入,进行迭代优化,同时使用判别模型S来检测图像是否存在隐写操作,反馈生成图像质量的特性,G&D&S三者在迭代过程中相互竞争,性能不断提高。该方案所采用的策略与SGAN(Steganographic GAN)和SSGAN(Secure Steganography based on GAN)两种方案不同,主要区别是将含密信息与真实图像样本作为判别模型的输入,对于判别网络D进行重构,使网络更好地评估生成图像的性能。与SGAN和SSGAN相比,该方案使得攻击者在隐写分析正确性上分别降低了13.1%和6.4%。实验结果表明,新的信息隐藏方案通过生成更合适的载体信息来保证信息隐藏的安全性,能够有效抵抗隐写算法的检测,在抗隐写分析和安全性指标上明显优于对比方案。

关键词: 信息隐藏, 隐写分析, 生成对抗网络

Abstract: Focusing on the issue that information-hidden carriers will leave traces of modification, and it is fundamentally difficult to resist statistical steganalysis algorithm detection, a new security steganography model based on Generative Adversarial Network (GAN) was proposed. In this scheme, the generator model G in GAN was utilized to generate the original carrier information with noise as the driver. Next, by using the ±1 embedding algorithm, the secret message was embedded into the generated carrier information to generate the secret information. Finally, the secret information and the real image sample were used as the input of discriminator D in the GAN for iterative optimization. At the same time, discriminative model S was used to detect whether the image has a steganography operation, and timely feedback to generate image quality features, G&D&S competed with each other in the iterative process, and the performance was continuously improved. The proposed strategy is different from the two schemes of Steganographic GAN (SGAN) and Secure Steganography based on GAN (SSGAN). The main feature is that the secret information and the real image sample are used as input for the discriminative model, and the discriminative network D is reconstructed, so that the network can better evaluate the performance of the generated images. Compared with SGAN and SSGAN, the proposed model reduces the detection accuracy of steganalysis by 13.1% and 6.4% respectively. Experimental results show that the new information hiding scheme guarantees the security of information hiding by generating more suitable carrier information and can effectively resist the detection of steganographic algorithms, it is significantly superior to the contrast schemes in terms of anti-steganography analysis and security indicators.

Key words: steganography, steganalysis, Generative Adversarial Network (GAN)

中图分类号: