数字视频帧间篡改和重压缩被动取证算法研究

发布时间：2018-05-28 05:34

本文选题：数字视频被动取证 + 帧间篡改　；参考：《北京交通大学》2017年博士论文

【摘要】：随着互联网和多媒体技术的蓬勃发展,数字视频应用范围越来越广泛。数字视频内容真实性问题日益严峻,许多功能强大的软件轻易编辑和篡改的视频效果十分逼真,公众无法简单判断视频内容的真伪。关系到社会公平和稳定的数字视频内容真实性的鉴定成为一个亟待解决的现实问题,数字视频取证技术应运而生。数字视频取证技术分为主动取证和被动取证两种,被动取证技术无需预先在视频载体中嵌入认证信息,仅需依靠视频自身数据特征可实现认证,具有更高的实用价值,成为近年来信息安全领域一个重要研究课题。在各种篡改手段中,数字视频帧间篡改容易操作、最为常见,因此对帧间篡改进行有效检测是数字视频被动取证领域的关键问题。另一方面,篡改视频必然经历过重压缩,重压缩检测是数字视频被动取证领域的研究热点。深入分析和研究视频帧间篡改取证技术和重压缩取证技术具有重要的理论意义和实际应用价值。本文的主要创新性工作包括以下四个方面。1.针对改善帧间篡改视频检测精度的问题,提出了两种视频帧间篡改取证算法。鉴于帧间篡改操作导致篡改点相邻两帧的图像内容相关性降低,构造了局部二值模式(Local Binary Pattern,LBP)相关系数的商特征和结构相似度均值(Mean of Structural Similarity,MSSIM)的商特征来表征图像内容相关性的变化,利用切比雪夫不等式和阈值法检测异常峰值,实现了对帧间篡改视频的检测及篡改点定位。2.针对视觉难以鉴别的帧间篡改视频的检测精度问题,构造了运动矢量金字塔(Motion Vector Pyramid,MVP)特征和后处理MVP特征,提出了两种视频帧间篡改取证算法。MVP将运动矢量模块和图像金字塔相结合抓住视频相邻两帧之间内容的细微变化,精确地反映了两帧之间的运动关系。再对MVP特征进行去均值、累加、直方图统计的后处理,充分利用帧间篡改遗留的痕迹,提高了算法对原始视频、帧删除视频和帧复制视频的分类能力。3.针对数字视频被动取证领域中相同量化参数下的重压缩取证问题,提出了一种相同量化参数下的H.264/AVC视频多重压缩取证算法。以相邻三次压缩间不同量化DCT系数的比率差构建了含有四分位数的特征集,作为支持向量机的输入,实现了对单次压缩视频和多重压缩视频的分类。实验结果表明,所提算法有高分类精度,对复制/粘贴攻击和帧删除攻击具有较强的鲁棒性。4.针对最新一代视频编码标准HEVC的重压缩取证问题,提出了一种不同码率下的HEVC视频重压缩取证算法。将HEVC视频中Ⅰ帧预测单元(Prediction Unit,PU)划分类型的块数目特征和量化DCT系数的首位数字概率分布特征结合构成融合特征,更全面地反映了重压缩对视频数据的影响,更好地判定视频的压缩历史。实验结果表明,所提算法能有效区分HEVC单次压缩视频和HEVC双重压缩视频。论文以篡改点突变特性为基础,从构造反映图像内容信息的LBP相关系数商特征和MSSIM商特征延展到构造反映两帧之间内容细微变化的MVP特征,结合切比雪夫不等式和广义ESD检验对视频帧间篡改取证算法进行了研究。另外以压缩编码有损特性为基础,从广泛使用的H.264/AVC编码标准的重压缩取证扩展到最新编码标准HEVC的重压缩取证,为视频重压缩取证技术的研究提供了多种思路。
[Abstract]:With the rapid development of Internet and multimedia technology, the application range of digital video is becoming more and more extensive. The problem of the authenticity of digital video content is becoming more and more serious. Many powerful software editors and tampering are very realistic. The public can not simply judge the authenticity of the video content. It is related to the social fairness and stability of the digital view. The identification of the authenticity of frequency content has become a real problem to be solved urgently. The digital video forensics technology emerges as the times require. The digital video forensics technology is divided into two kinds of active forensics and passive forensics. The passive forensic technology needs not to embed the authentication information in the video carrier in advance. It only needs to rely on the feature of the video's own data to achieve the authentication. The practical value has become an important research topic in the field of information security in recent years. In various tampering means, digital video interframe tampering is easy to operate and is most common. Therefore, effective detection of interframe tampering is the key problem in the field of digital video passive forensics. The other side, tampering video must undergo heavy compression and pressure shrinkage check. Testing is a hot topic in the field of digital video passive forensics. It is of great theoretical significance and practical application value to analyze and study video interframe forensics and recompression technology. The main innovative work of this paper includes the following four aspects: the problem of improving the accuracy of video interframe tamper modification by.1., two Since the interframe tamper operation leads to the decrease of the correlation of two adjacent frames, the quotient features of the local two value mode (Local Binary Pattern, LBP) correlation coefficient and the quotient feature of the structural similarity degree (Mean of Structural Similarity, MSSIM) are constructed to characterize the correlation of the content of the image. By using Chebyshev inequality and threshold method, the detection accuracy of interframe tamper video detection and tamper location.2. is realized. The motion vector Pyramid (Motion Vector Pyramid, MVP) features and post-processing MVP features are constructed, and two kinds of video are proposed. Inter frame tampering and forensics algorithm.MVP combines motion vector module and image Pyramid to capture the subtle changes in the content between two frames of adjacent frames and accurately reflects the motion relationship between the two frames. Then the MVP features are processed by means of averaging, accumulating, histogram statistics, and making full use of the traces left by interframe tampering to improve the algorithm pair The classification ability of original video, frame deleting video and frame copy video.3. aims at the problem of heavy compression and forensics under the same quantization parameters in the field of digital video passive forensics. A multiple compression proof algorithm for H.264/AVC video under the same quantization parameters is proposed. The ratio difference quantization of DCT coefficients between adjacent three times compression is constructed. With four quantile feature sets, the classification of single compression video and multiple compressed video is implemented as the input of support vector machine. The experimental results show that the proposed algorithm has high classification accuracy, strong robustness to copy / paste attacks and frame deleting attacks, and.4. for the latest generation video coding standard HEVC In this paper, a HEVC video recompression algorithm under different bit rate is proposed, which combines the block number feature of Prediction Unit (PU) in HEVC video and the first digital probability distribution characteristic of quantized DCT coefficient to form the fusion feature, and the effect of heavy compression on video data is reflected more comprehensively, and it is better judged. The experimental results show that the proposed algorithm can effectively distinguish HEVC single compression video and HEVC dual compression video. Based on the mutation characteristics of tamper points, the paper constructs the MVP features of the LBP correlation coefficient quotient features and the MSSIM quotient features that reflect the content information of the image to the MVP features of the subtle changes in the content between the two frames of the antireflection frame. Combining the Chebyshev inequality and the generalized ESD test, the video interframe forensics algorithm is studied. On the basis of the lossy feature of the compressed coding, the heavy compression evidence from the widely used H.264/AVC coding standard is extended to the recompression of the latest coding standard HEVC, which provides a variety of research for the video frequency recompression technology. Thinking.
【学位授予单位】：北京交通大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】