通信学报 ›› 2021, Vol. 42 ›› Issue (9): 144-154.doi: 10.11959/j.issn.1000-436x.2021174

• 学术论文 • 上一篇    下一篇

基于轮廓自动生成的构造式图像隐写方法

周志立1,2, 王美民1,2, 杨高波3, 朱剑宇1,2, 孙星明1,2   

  1. 1 南京信息工程大学数字取证教育部工程研究中心,江苏 南京 210044
    2 南京信息工程大学计算机与软件学院,江苏 南京 210044
    3 湖南大学信息科学与工程学院,湖南 长沙 410082
  • 修回日期:2021-08-09 出版日期:2021-09-25 发布日期:2021-09-01
  • 作者简介:周志立(1984− ),男,湖北黄冈人,博士,南京信息工程大学教授,主要研究方向为信息隐藏、数字取证、视觉密码、数字多媒体内容安全等
    王美民(1996− ),男,江苏盐城人,南京信息工程大学硕士生,主要研究方向为信息隐藏、数字取证
    杨高波(1974− ),男,湖南岳阳人,博士,湖南大学教授,主要研究方向为图像/视频信号处理、多媒体通信、数字媒体内容安全等
    朱剑宇(1996− ),男,江苏南通人,南京信息工程大学硕士生,主要研究方向为数字水印、图像处理
    孙星明(1963− ),男,湖南湘潭人,博士,南京信息工程大学教授,主要研究方向为网络与信息安全、传感器网络、自动气象观测等
  • 基金资助:
    国家重点研发计划基金资助项目(2018YFB1003205);国家自然科学基金资助项目(61972205);国家自然科学基金资助项目(61972143)

Generative steganography method based on auto-generation of contours

Zhili ZHOU1,2, Meimin WANG1,2, Gaobo YANG3, Jianyu ZHU1,2, Xingming SUN1,2   

  1. 1 Engineering Research Center of Digital Forensics, Ministry of Education, Nanjing University of Information Science and Technology, Nanjing 210044, China
    2 School of Computer and Software, Nanjing University of Information Science and Technology, Nanjing 210044, China
    3 College of Computer Science and Electronic Engineering, Hunan University, Changsha 410082, China
  • Revised:2021-08-09 Online:2021-09-25 Published:2021-09-01
  • Supported by:
    The National Key Research and Development Program of China(2018YFB1003205);The National Natural Science Foundation of China(61972205);The National Natural Science Foundation of China(61972143)

摘要:

为解决现有构造式隐写方法隐藏容量小和秘密信息难以提取的问题,提出一种基于轮廓自动生成的构造式图像隐写方法,具体包括以秘密信息为驱动的轮廓线生成和从轮廓线到图像变换2个过程。首先,建立基于长短期记忆网络(LSTM)的轮廓自动生成模型,实现以秘密信息为驱动的图像轮廓线生成;然后,建立基于pix2pix模型的轮廓-图像可逆变换模型,将轮廓线变换为含密图像。该模型也支持含密图像到轮廓的逆变换,从而实现秘密信息提取。实验结果表明,所提方法不仅能有效地抵抗隐写分析攻击,还能实现较高的隐藏容量和准确的秘密信息提取,性能明显优于现有的同类构造式图像隐写方法。

关键词: 构造式图像隐写, 无载体信息隐藏, 深度学习, 生成对抗网络

Abstract:

To address the problems of limited hiding capacity and inaccurate information extraction in the existing generative steganography methods, a novel generative steganography method was proposed based on auto-generation of contours, which consisted of two main stages, such as the contour generation driven by secret information and the contour-to-image transformation.Firstly, the contour generation model was built based on long short term memory (LSTM) for secret information-driven auto-generation of object contours.Then, a contour-to-image reversible transformation model was constructed based on pix2pix network to obtain the stego-image, and the model also supported the reversible transformations from the stego-image to contours for secret information extraction.Experimental results demonstrate that the proposed method not only achieves high hiding capacity and accurate information extraction simultaneously, but also effectively resists the attacks by steganalysis tools.It performs much better than the state-of-the-art generative steganographic methods.

Key words: generative steganography, coverless information hiding, deep learning, generative adversarial network

中图分类号: 

No Suggested Reading articles found!