【作者】 宣曼；

【导师】 蒋建国；

【作者基本信息】 合肥工业大学 ， 信号与信息处理， 2011， 博士

【摘要】 近年来，数字信号处理技术和宽带网络设施的飞速发展，使得数字内容的制造和分发越来越便利。曾经只有专业水准才能完成的数字图像/视频的编辑工作，如今可以借助很多便宜甚至免费的软、硬件轻松完成。这种便利无疑为大多数内容用户带来方便和实惠，但也会产生一系列问题。首先，在不影响合法用户享受数字多媒体的前提下，对诸如数字音乐，电影和书籍等多媒体内容进行版权保护，成为了一项新的课题，亟待研究。其次是内容认证方面，由于多媒体数据的易编辑特性，使得人们对它的真实性产生质疑，特别是当这些数据用作呈堂证供的时候。此外，为了保护公民财产安全，很多公共场合都安装监控摄像机。为了方便存储和传输，越来越多的监控摄像机采用数字格式存储视频。监控系统的视觉数据无疑是很重要的，并且他们的真实性需要得到保证，这样数字视频才能够无可置疑地作为可靠和有效的证据使用。因此，编码传输过程中鲁棒、高效的视频认证及篡改修复机制的研究成为信息安全领域的热点问题。针对上述诸多问题，本课题重点在H.264/AVC编码标准的基础上，结合视频序列的时间特性与编码特性，设计高效、鲁棒，水印容量可调的认证算法，对H.264编解码过程中的码长、码率等指标的影响降低到最小，同时进行篡改检测、定位机制与篡改恢复机制的研究，形成一套完善的视频认证系统。主要研究成果包括：1、基于残差块量化分层系数模型与“十字区域”运动矢量相结合的H.264鲁棒水印算法。视觉质量是水印算法性能最直观的衡量标准，为了提高算法的视觉质量，通常需要对图像进行复杂度分析。在压缩编码传输过程中的视频水印算法，比图像水印算法增加了时间维度的信息，即大量的帧间信息。简单的图像复杂度分析不再适用于视频水印算法。本课题着力针对亮度、纹理、空间信息综合考虑的视频复杂度分析，通过调制帧内目标宏块的残差量化分层系数与帧间宏块“十字区域”的运动矢量来共同完成标识信息的嵌入，有效缓解视频质量与水印算法鲁棒性之间的矛盾。2、基于零系数备用空间与帧间预测模式相结合的H.264容量多级提升的信息隐藏算法。水印容量是衡量水印算法性能的另一重要标准，扩大水印容量对于传统的水印算法而言，势必会使视频质量下降，如何在水印容量与视频质量之间进行取舍是水印算法的关键问题。本课题将这方面作深入研究，帧内编码部分利用DCT变换、量化后残差块的若干零系数构造备用空间用来存储标识水印信息；在帧间编码部分利用H.264帧间可变长码（VLC）的无差值补偿特性，引入信息变长分组概念，在帧间预测模式部分隐藏标识水印信息，本文将帧内帧间两部分相结合设计一种安全的水印容量可多级提升的视频信息隐藏算法，有效改善视频质量与水印容量之间的矛盾。3、基于数字水印技术的H.264主动目标识别算法。智能视频监控系统包括运动目标分割、目标识别以及目标跟踪等环节的研究。其中目标识别主要是在运动目标分割、提取的基础上，对每个目标（监控范围内有多个目标）提取其自身的不变特征，做出标记，以绘制出各目标的运动轨迹，是智能视频监控的主要内容。本文提出基于水印的主动目标识别概念，在H.264标准上，当前帧中进行运动目标检测并分割，在重构帧的运动目标中嵌入不同的水印，重构图像将在下一帧用于预测，这样水印信息被传递到下一帧，在下一帧与中通过提取参考帧运动目标预测值中的水印信息即可有效地实现目标识别，随后继续在下一帧的重构帧中嵌入水印，循环处理，直至完成整段视频的运动目标识别。4、H.264视频完整性认证系统及篡改恢复机制研究。视频完整性认证系统应当包括认证码的提取、标识信息的嵌入，篡改检测及定位几个过程，视频序列比图像信息更复杂，不能简单的采用图像hash值提取的方法提取视频认证码。本课题提出空间能量波动分析方法，针对不同特征的视频帧，有针对性的采用改进的SQ与VQ量化方法，有效提取出视频认证码，并结合改进的前文中H.264鲁棒水印算法与H.264大容量信息隐藏算法，构成H.264视频完整性认证系统，并进行篡改修复策略的研究，在一定程度上修复篡改操作。更多还原

【Abstract】 In recent years, digital signal processing technology and broadband network infrastructure havedeveloped rapidly, which makes the manufacture and distribution of digital content more and moreconvenient. In the past only the professionals could edit the multimedia files, the work that can beaccomplished easily by some cheap or even free software and hardware nowadays. The change ofsituation brings convenience and benefits to most of the content users undoubtedly, however, somepotential problems come into being. First of all, a new issue is to exert: how to carry out copyrightprotection for digital multimedia contents (eg, digital music, movies and books) without affectinglegitimate users’ normal operation? Secondly, concerning the content authentication, the authenticityof multimedia data will be questioned due to its easily editing feature, especially when those data areused as sworn evidence in court. In addition, many surveillance cameras are installed in publicplaces to protect the civil property. In order to facilitate storage and transmission, digital videoformat is applied in more and more surveillance cameras. Visual data of guaranteed authenticity insurveillance system, with which digital video can be served as reliable and valid evidence, which isvery important. Therefore, research on robust coding transmission, efficient video authentication andtamper recover mechanism becomes a hot issue in the field of information security.In response to problems mentioned above, this paper based on the H.264/AVC coding standard,mainly focus on designing an efficient, robust and adjustable watermark capacity authenticationalgorithm on the combination of time and coding characters of video sequences. The algorithmminimizes the influence on the parameter indicators such as bit-length and bit-rate in the H.264coding process. At the same time, researches on the tamper detection mechanism and tamperrecovery mechanism form a set of video authentication system. The main achievements are listed asfollows.First of all, a novel robust watermarking algorithm is proposed, based on the quantized layeredcoefficients of the residual macroblock and the motion vectors in the “crossing region”.Visual quality is the most intuitive standard measurement of watermarking algorithmperformance. In order to improve the visual quality of watermarking algorithm, image complexityanalysis is always necessary. Compared with the image watermarking algorithm, videowatermarking algorithm in the coding and transmission includes the information of time dimension,namely a large number of inter-frame information. Simple analysis of image complexity no longerapplies to video watermarking. The study emphasizes on the video complexity analysis composed ofbrightness, texture, spatial information, and embedding watermarking by modulating the quantized layered coefficients of intra residual macroblock and motion vectors in the “crossing region” of theinter-frame macroblock. The algorithm effectively alleviates the conflict between video quality androbustness of video watermarking algorithm.Secondly, the H.264video information hiding algorithm with adjustable watermarking capacityis proposed, based on spare space formed by zero-value DCT coefficients and inter prediction mode,.Watermarking capacity is another standard measurement of watermarking algorithmperformance. To the traditional algorithm, expanding the quantity of watermark embedded willdegrade the visual quality. Thus the key issue of watermarking algorithm lies in the trade-offbetween watermark capacity and video quality. The further research launches in this area. During theintra-frame coding, quantized zero-value DCT coefficients of residual marcoblock are utilized toform a spare space, information hiding is achieved by modifying one zero-value DCT coefficients ofspare space. In the course of inter-prediction, we introduce the concept of information variablelength group and achieve information hiding by using no difference compensation characteristics ofthe H.264inter-frame variable length coding. These two courses are composed to form H.264videoinformation hiding algorithm with adjustable watermarking capacity, improving the conflict betweenwatermarking capacity and visual quality of video watermarking algorithm.Thirdly, H.264active object recognition is proposed based on digital watermarking.Intelligent video surveillance system includes the study on moving object segmentation, objectrecognition, object tracking and so on. Among these, based on the results of object segmentation,object recognition is the main content of intelligent video surveillance system, which extracts theinvariant features of each object, making a mark to draw each one’s trajectory. In this dissertation,we proposed the concept of active object recognition based on digital watermarking. On the H.264coding standard, we detect the moving objects and segment them. Then we embed differentwatermarks in different objects in the reconstructed frame, which is used as the reference frame forprediction in next frame, thus, the watermarks are transmitted to the next frame. Object recognitionis realized by extracting the watermarks from the objects in the reference frame of the next frame,and embedding the watermarks in objects in the reconstructed image of the next frame, doing thisprocess circularly until completing the whole moving object identification.Finally, H.264/AVC video integrity authentication system is proposed and the research ontamper recovery mechanism is carried.Video integrity authentication system includes the processes of authentication code extraction,identification information embedding, tamper detection and localization. Video is more complicatedthan image, so that the image hash extraction method is not suitable to extract video authenticationcode. The analysis method of spatial energy fluctuation is proposed, aiming to improve the SQ orVQ quantitative methods to extract the video authentication codes effectively from video frames with different characteristics. Then we constitute H.264video integrity authentication system whichconsists of improved robust watermarking algorithm and information hiding algorithm with largecapacity aforementioned. At last, the research carried on tamper recovery mechanism to recover thetamper area in a certain degree.更多还原