基于自监督学习PBS-Net和通道提纯的信息隐藏主动防御方法  

Active defense method for information hiding based on self-supervised learning PBS-Net and channel purification

作　　者：马媛媛[1,2] 赵颖澳 徐富永 张倩倩[1,2] 辛现伟 Ma Yuanyuan;Zhao Ying’ao;Xu Fuyong;Zhang Qianqian;Xin Xianwei(College of Computer&Information Engineering,Henan Normal University,Xinxiang Henan 453007,China;Engineering Lab of Intelligence Business&Internet of Things,Henan Normal University,Xinxiang Henan 453007,China)

机构地区：[1]河南师范大学计算机与信息工程学院,河南新乡453007 [2]河南师范大学智慧商务与物联网技术河南省工程实验室,河南新乡453007

出　　处：《计算机应用研究》2024年第12期3822-3828,共7页Application Research of Computers

基　　金：国家自然科学基金资助项目(62002103);河南省优秀青年科学基金资助项目(222300420058);河南省高等学校重点科研项目(24A520019)。

摘　　要：信息隐藏主动防御技术作为信息隐藏的对立面,能够阻断非法隐蔽通信的传输。然而,现有的主动防御方法过度依赖载体-载密图像对,无法对未知载密图像主动防御,使其防御的误码率在实际社交网络中降低。针对上述问题,为了在通信双方毫无察觉的情况下彻底阻断秘密信息的传输,提出一种自监督学习盲点网络和通道提纯的主动防御方法。首先,通过像素混洗采样策略降低载密图像中像素之间的空间相关性,将学习方式从监督学习改进为自监督学习;其次,中心掩码卷积和空洞卷积残差块用于消除载密图像中的秘密信息;最后,设计通道提纯模块改善图像纹理细节。该方法无须任何信息隐藏方案的先验知识以及人工操作,使得在主机接收到可疑图像之前消除秘密信息,阻断社交网络中的隐蔽通信。实验结果表明,该方法具有高秘密信息破坏效果和高图像质量,能够达到100%的防御成功率,阻断社交网络中的隐蔽通信。同时,在不同负载率的数据集下,该方法与SC-Net和AO-Net进行对比,在秘密信息消除方面各提升14.14%和2.91%,在图像质量方面各提升9.14%和43.34%。As the opposite of information hiding,active defense technology can block the transmission of illegal hidden communication.However,the existing active defense methods rely too much on cover-stego image pair and cannot actively defend the unknown stego images,which reduce the bit error ratio in the actual social network.To solve these problems,this paper proposed an active defense method of self-supervised learning blind-spot network and channel purification in order to completely block the transmission of secret information without being noticed by both communication parties.Firstly,the method used pixel shuffling sampling to reduce the spatial correlation of pixels in stego images,and improve the learning mode from supervised learning to self-supervised learning.Secondly,it integrated centrally masked convolutions and dilated convolution residual blocks to eliminate secret information.Finally,it obtained the channel purification module to improve the image texture details.The method didn’t need any prior knowledge of information hiding schemes and manual operation,so that it could eliminate secret information before hosts receive suspicious images.The experimental results show that this method has high secret information destruction effect and high image quality,and can achieve 100%defense success rate and block covert communication in social networks.At the same time,under different payload data sets,the proposed method is compared with SC-Net and AO-Net,and the secret information elimination is improved by 14.14%and 2.91%respectively.The image quality was improved by 9.14%and 43.34%respectively.

关 键 词：图像隐写分析 主动防御 自监督学习 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]