Journal of Information Security Reserach ›› 2023, Vol. 9 ›› Issue (6): 566-.

Previous Articles     Next Articles

Research on Image Steganography and Extraction Scheme Based on  Implicit Symmetric Generative Adversarial Network

  

  • Online:2023-06-04 Published:2023-06-03

基于隐式对称生成对抗网络的图像隐写与提取方案

屈梦楠, 靳宇浩, 邬江   

  1. (中电长城网际安全技术研究院(北京)有限公司北京100097)
  • 通讯作者: 屈梦楠 工程师.主要研究方向为图像处理与机器学习. quanfita98@163.com
  • 作者简介:屈梦楠 工程师.主要研究方向为图像处理与机器学习. quanfita98@163.com 靳宇浩 工程师.主要研究方向为网络与信息安全、威胁情报. jyh8888@88.com 邬江 博士研究生,高级工程师.主要研究方向为网络与信息安全. wujiang@cecgw.com

Abstract: Aiming at the problems in the image steganography technology that the quality of the carrier image is degraded and vulnerable to attacks when the secret image is embedded, this paper proposes an image steganography and extraction scheme based on an implicit symmetric generative network. The scheme first abstracts the task of image steganography and extraction into a mathematical optimization problem. Secondly, an implicit symmetric generative adversarial network model is proposed according to the optimization problem. The implicit symmetric generative adversarial network contains two independent generative adversarial subnetworks, namely the steganographic adversarial subnetwork and the extraction adversarial subnetwork. In the steganographic confrontational subnetwork, first the encoder converts the cover image and the covert image into a set of highdimensional feature vectors containing enough cover image information and secret image information. The decoder then reconstructs these feature vectors into images embedded with secret information. In the extraction adversarial subnetwork, the image embedded with secret information is passed through another set of encoder and decoder to extract the hidden image. Finally, a loss function suitable for the model is designed. Experimental results show that the proposed scheme has high image quality and can maintain good robustness in the face of various common attacks.

Key words: image steganography, generative adversarial network, privacy protection, symmetric generative network, image to image

摘要: 针对图像隐写技术中存在嵌入秘密图像时载体图像质量下降、易受攻击等问题,提出一个基于隐式对称生成对抗网络的图像隐写与提取方案.该方案首先将图像隐写与提取任务抽象为一个数学优化问题.其次,根据该优化问题提出一个隐式对称生成对抗网络模型.在隐式对称生成对抗网络中包含2个相互独立的生成对抗子网络,即隐写对抗子网络和提取对抗子网络.在隐写对抗子网络中,首先编码器将载体图像和隐秘图像转换为1组包含足够多的载体图像信息和秘密图像信息的高维特征向量,之后解码器将这些特征向量重新构造为嵌入秘密信息后的图像.在提取对抗子网络中,将嵌入秘密信息的图像通过另一组编码器和解码器提取出隐秘图像.最后,设计适用于该模型的损失函数.实验结果表明,该方案具有较高的图像质量,并且能够在面对各种常见攻击时保持较好的鲁棒性.

关键词: 图像隐写, 生成对抗网络, 隐私保护, 对称生成网络, 以图藏图