基于率失真性能检测的视频隐写分析技术研究

Digital video is one of the most influential media, and widely believed to be an ideal cloak for covert communication. Thus video steganography and video steganalysis have become the important research topics in the field of information hiding. Compared with image steganalytic techniques which are increasingly sophisticated, video steganalytic techniques are still in an early stage of development. First, there is a lack of targeted video steganalytic methods for certain types of embedding domains. Second, the existing video steganalytic methods have a fairly limited detection range, and always lose their effectiveness when faced with multiple embedding domains. Besides, they are vulnerable to cover source mismatch. To address the limitations and shortcomings of the current video steganalytic techniques, the evaluation of rate-distortion (RD) performance is considered in this project because RD optimization is usually performed in video coding and RD performance of video coding would be adversely affected by embedding operations. This project is aimed at proposing highly effective video steganalytic methods and the following measures are taken. First, mathematical models for evaluating RD performance are developed by performing accurate measurement of visual distortion and compression efficiency. Second, reasonable rules are established to check whether or not RD performance of video coding is optimal. Further, some other effective mechanisms for video steganalysis are carefully considered in the feature engineering. This project could facilitate the development of video steganographic and steganalytic techniques, and can provide solid support for government in censorship of illegal use of video steganography.

数字视频作为当前最具影响力的传播媒介，被普遍视为理想的隐蔽通信载体，视频隐写和视频隐写分析也因此成为信息隐藏领域的研究热点。相比日益成熟的图像隐写分析技术，视频隐写分析技术尚处于较为初级的发展阶段，存在技术体系不够健全、检测范围有限、鲁棒性较低等局限性。针对此现状，本项目将根据视频编码通常采用了率失真优化技术这一事实，利用隐写操作将不可避免地破坏视频编码的率失真性能这一本质弱点，研究基于率失真性能检测的视频隐写分析技术。具体地，本项目将通过建立合理的率失真性能计算模型和精确的率失真性能最优性判定准则，并优选融合其他隐写分析特征设计机制，从而为高性能视频隐写分析方法的设计提供解决方案。本项目的实施，将有助于推动视频隐写和视频隐写分析技术的发展，并为国家相关部门在视频隐写通信的监管方面提供有力支撑。

视频隐写分析技术用于检测数字视频中是否存在隐写修改痕迹，是信息隐藏领域的研究热点。开展相应研究不仅可支撑国家有关部门对视频隐写通信的监管，还有助于推动视频隐写和视频隐写分析技术的发展。针对当前视频隐写分析技术存在的体系不健全、检测范围有限、鲁棒性较低等局限性，本项目围绕以下三方面展开研究：1）视频码流语法元素率失真性能计算模型的建立；2）视频码流语法元素率失真性能最优性判定准则的构建；3）高性能视频隐写分析特征的设计。本项目的实施取得了以下三方面重要结果：1）针对预测模式、量化参数等状态变化范围较小的码流语法元素，分别提出了合理的率失真性能最优性判定准则，在此基础上设计了有效的专用隐写分析特征；2）对变换系数进行调制修改会对环路去块滤波、预测残差差分、熵编码码字统计特性造成扰动，基于此，提出了多种针对变换系数域隐写的专用分析方法；3）孪生卷积神经网络有助于构建适用于通用视频隐写分析任务的基本检测框架。本项目在一定程度上突破了视频隐写分析领域的技术瓶颈，部分成果已发表于领域重要期刊及会议，成果系统已在应用单位得到部署与使用。