【作者】 徐长勇；

【导师】 平西建；

【作者基本信息】 解放军信息工程大学 ， 军事情报学， 2009， 博士

【摘要】 信息隐藏是互联网时代信息安全领域的重要研究内容之一。作为信息隐藏技术的重要分支,数字隐写研究如何将秘密信息以不引起第三方注意的方式隐藏到公开的数字媒体中,并在公共信道中进行传输,亦即隐蔽通信问题。与数字隐写相对抗的技术被称作隐写分析,通过分析来检测被怀疑对象中是否存在隐藏信息。视频是互联网上常见的数字媒体形式,与图像等媒体形式相比,视频由一系列的帧序列组成,具有更大的可用载体空间。视频压缩编码技术和网络流媒体业务的发展,使得视频的应用范围越来越广,如视频点播、视频会议,以及随着家用摄像机的日益普及和各种操作简便的视频编辑软件的应用,人们可以方便的进行视频的录制、剪辑并在网络上发布、交流,从而使得网络上的视频传输越来越常见,因此有必要开展视频数字隐写以及与之相对抗的隐写分析技术的研究。本文针对信息安全的技术发展和应用需求,在深入分析视频数据特点的基础上,研究以压缩视频为载体的数字隐写方法以及针对具体隐写方法的隐写分析。本文的工作和取得的研究成果主要包括以下四个方面。1.基于运动矢量的压缩视频数字隐写算法研究:针对压缩视频流中存在大量运动矢量数据的特点,提出一种以压缩视频为载体的基于运动矢量的数字隐写算法。该算法在对典型的基于运动矢量的信息嵌入方法进行分析的基础上,按照GOP的顺序,采用冗余嵌入和控制信息嵌入相结合的方法,在帧间编码帧的运动矢量中以冗余嵌入的方式重复嵌入要传输的秘密信息,并在每帧中嵌入索引信息以对抗帧同步攻击,同时将与冗余嵌入有关的信息作为控制信息嵌入到对帧内编码帧进行编码的DCT系数中。实验结果表明,该算法在保持载密视频质量的前提下,能够有效抵抗帧添加、帧删除等帧同步攻击。2.基于纠错码的压缩视频数字隐写算法研究:在对MPEG压缩视频数据进行分析的基础上,针对在采用差分编码的系数中嵌入信息后还需进行差值补偿的特点,提出了一种在压缩视频流中不采用差分编码的DCT系数中嵌入信息的数字隐写算法。该算法根据纠错码数字隐写的原理,采用二次嵌入的策略,首先将秘密信息嵌入纠错码码字,然后将得到的载密纠错码码字嵌入到与没有采用差分编码的系数相对应的行程幅度对的幅度值中,并在秘密信息嵌入过程中采用视频流长度控制策略。实验结果表明,该算法在确保一定隐藏容量的前提下,具有较好的视觉不可感知性和统计不可感知性,同时能够使得秘密信息嵌入前后的视频流长度近似保持不变。3.针对视频隐写软件MSU StegoVideo的隐藏信息检测与主动攻击方法研究:视频信息隐藏软件MSU StegoVideo采用了一种具有较强鲁棒性的嵌入算法,可以将多种类型的其它文件隐藏到AVI格式视频中,即使对载密视频进行二次压缩,仍能以较高准确率将秘密信息提取出来。在对该软件的隐藏原理进行分析的基础上,在唯载密攻击的条件下,根据嵌入信息后的数据分布特点,提出了基于块效应分布不均匀性和边界不连续性差异的隐藏信息检测方法,以及基于相邻帧帧间共谋的主动攻击方法。实验结果表明,隐藏信息检测方法能够实现对视频中隐藏信息存在性的有效检测,主动攻击方法能够在不影响视频质量的前提下,有效去除利用该软件嵌入的隐藏信息。将两种方法相结合,可以检测并破坏利用MSU StegoVideo实施的隐蔽通信。4.基于时空相关性的视频隐写分析方法研究:时间相关性和空间相关性是视频序列的基本属性,针对视频中的扩频隐写算法,将隐写过程建模为加性高斯噪声的叠加过程,分析了噪声叠加对时间相关性和空间相关性的影响,提出了基于时空相关性的隐写分析方法。该方法利用视频序列中各帧的水平、垂直、正对角和反对角四个方向的差分直方图度量隐写对空间相关性的影响,同时利用相邻帧帧间差分的直方图度量隐写对时间相关性的影响,然后根据隐写前后的差分直方图变化,使用从直方图获得的统计参数作为特征向量来辨识隐写前后的时空相关性变化,最后利用SVM对特征向量进行训练和分类,从而达到隐写分析的目的。实验结果表明,该方法对于未压缩视频和压缩视频,均能实现隐藏信息存在性的有效检测,同时空间相关性和时间相关性的有效结合,提高了检测性能。更多还原

【Abstract】 Information hiding is one of important research subjects in the field of information security. As a main branch of information hiding, steganography has received significant attention from both industry and academia. Steganography is the art of concealing the very existence of communications by embedding the secret information into the open digital media, such as digital image, digital video, and so on. The technique opposed to steganography is called steganalysis, the purpose of which is to expose the presence of secret information. Digital video is one kind of familiar digital media on the Internet. With the development of video compression technique and stream media operation, the application area of video is more and more widely, for example, video on demand and video conference. Moreover, the gradual prevalence of portable video camera and the video edit software, people can easily record, edit and release video on the Internet, which makes the video transmission more frequently. In addition, the characteristics of large data amount, abundant content and complicated statistics for video make it must be an important carrier for steganography. Consequently, it is of great importance to study the technique of video steganography and the opposed technique - video steganalysis.Keeping up with the newly developed techniques of steganography and staganalysis, and according to the analysis to video data, this thesis focus on the techniques of video steganography and video steganalysis. The main contributions of this thesis are summarized as follows:1. A steganographic method in compressed video stream based on motion vector. According to the fact that there are abundant motion vectors in video stream, a steganographic method based on motion vector is proposed. On a GOP by GOP basis, the strategy of redundant embedding and control information embedding is utilized to resist frame synchronization attacks. The secret information is embedded into the motion vectors of inter-encoded frames by redundant embedding firstly, then the control information generated during redundant embedding is embedded into the DCT coefficient of intra-encoded frames. Experimental results show the method can keep the quality of stego video, and resist the temporal synchronization attacks as frame inserting and frame dropping.2. A steganographic method in compressed video stream based on error-correction code. According to the thorough analysis to MPEG compressed video data, a steganographic method for embedding secret information in DCT coefficients which are not differential encoded is proposed. Utilizing the principle of steganography based on error-correction code, a strategy of two-time embedding is designed to embed secret information. That’s to say, the secret information is firstly embedded into the error-correction codeword, then by modifying the level value of run-level pair corresponding to the DCT coefficient which is not differential encoded, the stego codeword obtained by first time embedding is secondly embedded. Moreover, measures are taken for the purpose of preserving the video stream size. Experimental results show the method has a good visual and statistical imperceptibility, and can keep the nearly invariant size of video stream. 3. A hidden information detection method and an active attack method for MSU StegoVideo. MSU StegoVideo is an open video steganography software on the Internet, the embedding algorithm of which has a strong robustness. By the experiment of data embedding and extraction, the embedding mechanism is revealed and the distribution characteristic of the embedded data is analyzed. According to the analysis, on the condition of stego-only attack, a hidden information detection method utilizing the asymmetric distribution of block artifacts and the difference of edge discontinuity, and an active method based on inter-frame collusion is proposed. Experimental results show the detection method can accurately find the existence of hidden information, and the active attack can effectively destroy the hidden information on the premise that the video quality is preserved. The two methods can be combined to detect and destroy the covert communication which is executed by MSU StegoVideo.4. A video steganalysis method based on spatial-temporal correlation. Temporal correlation and spatial correlation are basic characteristics of video sequence. Aimed at spread-spectrum embedding for video which can resist many attacks such as video compression and noise addition, the data embedding is modeled as the addition of additive Gaussian noise. By analyzing the influence to temporal correlation and spatial correlation of video owing to noise addition, a steganalysis method based on spatial-temporal correlation is proposed. The method uses difference image histogram of four directions to measure the influence of spatial correlation, and temporal correlation is measured by the histogram of frame difference. According to the difference of the histograms after data embedding, the statistical parameters derived from the histograms are used as feature vectors to differentiate the change of spatial and temporal correlation. In the end, the stego video can be discriminated from the cover video by using SVM classifier to train and classify the feature vectors. Experimental results show it is effective to detect the existence of hidden information for both uncompressed video and compressed video. At the same time, it is testified that by the combination of spatial correlation and temporal correlation, the detection performance is improved.更多还原