【作者】 王好贤；

【导师】 孙圣和；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2008， 博士

【摘要】 数字水印技术是通过一定的算法将一些版权标志性信息直接嵌入到多媒体数据当中,但不影响数字多媒体的效果和使用价值,并且不能被人的知觉系统觉察或者注意到,只有通过专用的检测器才能抽取。数字水印最初研究的重点是数字图像水印,已经取得了不少的研究成果,而基于视频水印的技术和应用研究还相对较少,数字视频产品的版权保护亟需解决。视频水印算法大致可分为时空域、变换域和压缩域算法。传统的时空域算法在不可见性和安全性方面表现差,抗时间同步攻击能力差;传统的变换域算法意图将视频帧基于一组标准基进行分解,对帧图像来讲,这种分解不是帧图像的最佳表示;而压缩域算法在实时性和大容量上较难统一,本文针对上述问题提出了相应的算法。矢量量化技术是目前广为流行的多媒体数据的有损压缩技术,它已经被应用于图像压缩很多年了,本文率先将矢量量化应用于视频压缩编码,并针对编码方案设计了多功能水印算法。本文的主要内容和创造性研究成果如下:1.基于块分类图像水印算法,引入运动块的概念,提出了基于块分类的数字视频鲁棒水印算法。算法根据数字视频序列帧间运动信息和帧内细节信息对图像块进行分类,选择既属细节块又属运动块的图像块嵌水印,嵌入过程采用位平面替换的思想。最后通过实验证明了算法的有效性。2.在分析了DCT和SVD图像水印算法的基础上,提出了基于DCT域奇异值分解的数字视频鲁棒水印算法。算法首先将视频帧的Y分量分块进行DCT,选取直流和低频DCT系数进行奇异值分解,对奇异值系数量化嵌水印,并对所提算法进行了攻击实验。3.在分析了基于运动矢量视频水印算法的基础上,本文提出通过调整半像素精度运动矢量与整数精度运动矢量之间的夹角,在压缩过程中嵌水印的视频水印算法。另外,本文在分析可变长码域视频水印算法的基础上,通过调整相邻可变长码中DCT交流量化系数之间的关系嵌入水印,为了增加嵌入容量,对视频帧的色差分量也嵌入水印,仿真结果表明算法在满足实时性、大容量的同时对视频质量没有影响。4.基于矢量量化图像水印算法,提出基于矢量量化的多功能视频水印算法。本文结合矢量量化技术和H.263压缩标准,首先设计了基于空域矢量量化和变换域矢量量化的视频压缩方案。针对空域矢量量化视频压缩方案,算法利用I帧DCT直流系数的极性嵌入鲁棒水印,利用码字扩展技术在P帧预测差值矢量的量化索引中嵌入脆弱水印。针对变换域矢量量化视频压缩方案,利用码字索引标记技术分别将数字指纹和鲁棒水印嵌入I帧低频和中频DCT系数矢量索引中,利用索引受限技术将低频和中频矢量索引的信息嵌在高频矢量索引中。实验结果表明所提出的多功能水印算法是实用的。更多还原

【Abstract】 The technique of digital watermarking is to embed some marking information which claims the copyright in the digital multimedia data directly, the embedding process does not affect the effect and use value of the original content, also it can not be detected or noticed by human perception system. Furthermore, the embedded data can later be extracted or determined only through an appropriative watermark detector. Digital watermarking has first been extensively studied for still images, a lot of research results have been obtained, but the research of technique and applications for video watermarking is slower comparatively. Today, however, with the rapid development of multimedia technology, different kinds of video products are manufactured, as a consequence, the intellectual property protection problem of video products has become an urgent issue, and it is gaining increasingly recognition.In terms of the domain in which the watermark is inserted, video watermarking algorithms can be classified as spatial-temporal domain, transform domain or compressed domain watermarking algorithms. It is often claimed that embedding in the spatial-temporal domain is disadvantageous in terms of visibility and security, many of the spatial watermarking algorithms can’t resist temporal synchronization attack, the traditional transform domain watermarking algorithms attempt to decompose frames in terms of a standard basis set,but this is not necessarily the optimal representation for a given frame,also it is difficult for most of the previous work on compressed-domain video watermarking algorithms to achieve unification of real-time and big capacity. Aiming at the problems, this thesis puts forward corresponding algorithms.As we know, Vector quantization(VQ) is a popular lossy compression scheme widely used in multimedia data compressions,however,VQ has been used for image compression for many years, we first propose a video compression scheme based on VQ, we then propose a multipurpose watermarking algorithm for this video compression scheme.The main innovative contributions of this dissertation are as follows:Refering to block classification watermarking algorithm and introducing the concept of active block, this dissertation proposes a video watermarking algorithm based on block classification. First, the luminance of every whole frame is first divided into a number of non-overlapping blocks,then,different blocks of the original video are separated into perceptually distinct categories according to motion information and region complexity,the blocks are classified into detailed blocks or non-detailed blocks according to the energy of the DCT AC coefficients,besides, blocks are also classified into active blocks or inactive blocks according to the difference between the corresponding DC coefficients in the neighboring frames,what’s more,only blocks that lie in the intersection of detailed blocks and active blocks are selected for embedding. Meanwhile, the idea of bit-plane replacement is adopted in the embedding procedure. The simulation results verify the effectiveness of the algorithm.After analysing the image watermarking algorithm based on DCT and SVD, a watermarking method based on SVD in the DCT domain is proposed in this thesis. In our algorithm, the luminance of every whole frame is first divided into a number of non-overlapping blocks, then some blocks are randomly selected by a seed key according to watermark bit, besides, DCT is performed on the selected luminance block, the DC and low frequency components are selected to compose a coefficient matrix, what’s more, SVD is applied to each coefficient block, last but not least, the obtained singular values are quantized and modified according to the corresponding watermark bit.The experiment is done to test the algorithm against malicious attack.According to the study of video watermarking algorithm based on motion vector,this thesis proposes to embed watermark by adjusting the angle between the integer and half pixel accuracy motion vectors. In the same time, on the basis of analysis of video watermarking algorithms in variable length coding (VLC) domain, we propose another watermarking algorithm which is performed by adjusting the relationship of the neighbouring quantized AC coefficients in succeeded variable length code.In order to increase the embedding capacity, the chroma components of the video are also selected to embed watermark bits. Simulation results show that the proposed algorithms can perform the real-time watermark embedding process without affecting visual quality of the video sequence and it is also standard-compliant,also, the proposed algorithm can achieve big capacity. On the basis of the study of VQ image watermarking algorithms,two multipurpose video watermarking algorithms are proposed in this thesis. According to the combination of VQ and H.263 standard, two video compression scheme based on spatial domain vector quantization and transform domain vector quantization are designed.Aiming at the hybrid video compression scheme based on spatial domain vector quantization, the proposed watermarking algorithm embeds robust watermark by exploiting the polarities of DC (Direct Current) coefficients, and the codebook expansion technique is adopted to embed fragile watermark in VQ indices of the predicted differential vector. Aiming at the hybrid video compression scheme based on transform domain vector quantization,the code-words labeling technique is adopted to embed digital fingerprint and robust watermark in low-frequency and mid-frequency vector indices. The I-frame fragile watermarking algorithm embeds the information of low-frequency and mid-frequency vector indices in high-frequency indices. The P-frame fragile watermarking algorithm embeds the feature of motion vector in quantized DCT coefficients.The experimental results show that the multipurpose watermarking algorithms are available.更多还原