基于MPEG的实时视频水印研究-智能与分布计算实验室

基于MPEG的实时视频水印研究

姓名	凌贺飞
论文答辩日期	2005.05.08
论文提交日期	2005.05.13
论文级别	博士
中文题名	基于MPEG的实时视频水印研究
英文题名	Real-time Watermarking Techniques for MPEG-encoded Video Data
导师1	卢正鼎
导师2
中文关键词	数字水印;视频水印;实时性;大容量;几何攻击;差分数量水印;改进差分能量水印;能量调制水印
英文关键词	Digital watermarking;Video watermarking;Real-time;High capacity;Geometric attacks;Differential Number Watermarking (DNW);Improved Difference Energy Watermarking (IDEW);Energy Modulated Watermarking (EMW)
中文文摘	近年来，随着多媒体技术和通信网络的飞速发展，数字多媒体信息的存储、复制与分发变得非常方便。人们不但可以通过互联网和CD-ROM方便快捷地获得多媒体信息，还可以得到与原始数据完全相同的复制品。但是由此引发的盗版问题和版权纷争已成为日益严重的社会问题。数字水印技术作为解决这类问题的一种潜在的有效技术，正受到国内外学者的广泛关注。就公开的文献来看，绝大多数数字水印研究主要集中在静止图像上，而对视频水印研究甚少。随着大量数字视频产品如VCD、DVD 的出现，数字视频产品的版权保护更加迫切，越来越多的学者开始转移到视频水印研究上来。但是视频水印研究既是研究热点也是难点。与图像水印相比，视频水印需要更高的要求。一方面，视频水印算法在应用中经常有实时和接近实时的需求，所以其水印嵌入和提取算法必须具有较低复杂性。另一方面，由于受水印算法实时性的限制和视频水印自身的特殊要求，使得研究视频水印存在诸多难点，例如：在保证视频水印算法实时性的同时，如何抵抗几何失真问题，以及如何实现大容量水印嵌入等。在实时视频水印研究方面，尽管差分能量水印(DEW)算法本身表现出相当低的计算复杂性，但是给出的参数选择的优化算法计算复杂性过高, 同时鲁棒性比较差，需要进一步地提高。因此提出了改进差分能量水印(IDEW)算法，用能量差比率代替能量差，作为一比特水印信号的嵌入模式，使得强制能量差能够根据区域实际的整体能量作自适应的改变，可以增加嵌入的有效性和水印的鲁棒性。同时对嵌入过程中的截断索引的挑选进行改进，减少视觉失真。IDEW在水印容量、视觉失真和鲁棒性三方面上都比原始DEW算法好。并且算法的复杂性比基于统计模型的优化算法低得多。然而，DEW和IDEW都是运行在系数域中的算法，其计算复杂性相对较高。因此，又提出了一个直接运行在VLC域的实时水印算法??差分数量水印算法(DNW)。算法通过移动块结束标志(EOB)强制制造两配对区域中的元组数量差来实现水印比特位的嵌入。DNW算法的计算复杂性比标准VLC域水印算法的减少一半，与DEW算法和IDEW算法相比，其不需要量化过程，因此计算复杂性比DEW算法和IDEW算法低。并且DNW在水印的视觉失真、容量和鲁棒性方面都比DEW性能好。为了进一步提高算法的鲁棒性，又将Turbo码引入DNW算法中，提出了基于Turbo码的差分数量水印算法(T-DNW)。实验结果表明，Turbo码与DNW算法相结合，有利于提高算法的鲁棒性，但需要牺牲一半的有效载荷，同时增加了算法的计算复杂性。对于压缩视频的数字水印实际应用，特别是以压缩视频为载体的隐秘通信的水印应用，除了实时性要求外，水印容量特性也显得比较重要。因此提出了具有大容量、低复杂性的能量调制水印算法(EMW)，主要适合于(但不局限于) JPEG图像和MPEG视频流的实时水印嵌入与检测。主要通过在感知范围内调制块能量实现水印嵌入的。与多路径网格编码调制(MTM)方法相比，具有更低的计算复杂性。系统可实现在704×480的视频帧中嵌入2640个比特，水印容量相当可观，并且能够抵抗常见的处理和攻击，因此比较适用于大容量信息嵌入的实时视频水印的应用。抗几何攻击是数字水印领域所面临的最大困难，目前第一代数字水印和第二代数字水印方案都不能解决实时视频水印难题??既要抗几何攻击，又要满足视频水印的实时性需求。因此提出了一种抗几何攻击的实时视频水印新模型，该模型主要由部分解码过程、压缩域特征提取过程、局部区域水印嵌入过程和部分再编码过程等四部分组成。针对各部分的具体实现，给出了直接在视频压缩域中提取鲁棒特征点的特征提取方法；提出可调分辨率技术用于帧图像的快速解码与编码；提出更加实时鲁棒的水印嵌入与检测新方法，用于将水印信号重复嵌入到各局部区域。经分析和实验表明，新模型只需要将压缩视频进行部分解码，在整体上大幅度降低系统的计算复杂性，同时具备第二代数字水印的优点。
英文文摘	With the rapid development of multimedia technique and communication networks, the duplication and dissemination of multimedia data have recently become more and more convenient. Through Internet and CD-ROM, the digital copies with the same quality as original data can be got much more easily and quickly. But the consequential problems such as copyright piracy, illegal use and copyright dissension have increasingly become a serious social issue. Digital watermarking has been proposed to be a very useful technology in the copyright protection of digital data. The research has focused on still images for a long time but nowadays this trend seems to be changing. As a great deal of Digital Versatile Disks (DVD) flood the markets, the digital video data most needs protection, so more and more watermarking algorithms are proposed for video content. As a research hotspot, the research of video watermarking is also a difficult problem. Comparing with image watermarking, video watermarking should meet some additional specific requirements, such as real-time. In order to meet such a requirement, the complexity of the watermarking algorithm should obviously be as low as possible. Moreover, while keeping the real-time property of watermarking algorithms, how to resist geometric attacks and how to fulfill high-capacity watermark are another two difficulties. In recent research of real-time watermarking, the Difference Energy Watermarking (DEW) algorithm itself has been shown to have relatively low complexity, but the optimization approach for parameters setting is too complex to be practical, and the robustness is still low. So the Improved Difference Energy Watermarking (IDEW) algorithm was proposed, in which two measures are adopted. One is using the ratio of energy difference to total energy to replace energy difference as pattern to embed label bits, which makes the energy difference self-adapted according to the practical energy of an emdedding region, thus increase the watemark’s embedding validity and robustness. The other is modifying the selection of cut-off index in the watermark embedding process, which lowers visual quality impact. The experimental results show that the improved algorithm performs better on watermark’s visual quality impact, capacity and robustness than the original DEW algorithm. Moreover, its complexity is much lower than the optimization approach based on statistic model in the DEW algorithm. However, both IDEW and DEW algorithms run in the coefficient domain, and their complexities are a little higher than those in other domains. So a new real-time watermarking, Differential Number Watermarking (DNW) which can be directly performed in the VLC domain, has been proposed. The watermark bits are embedded in a pattern of number differences between block subsets by selectively removing high frequency components. The DNW algorithm has only half the complexity of other VLC domain watermarking algorithms, and comparing with the DEW and IDEW algorithm, it doesn’t need quantization step. So the DNW algorithm has relatively lower complexity, and performs better on watermark’s visual quality impact, capacity and robustness than the DEW algorithm. Furthermore, in order to improve the robustness, the Turbo-based DNW (T-DNW) algorithm was proposed in which the turbo code was employed in the DNW algorithm. From the analysis and experiments, it shows the T-DNW algorithm is much more robust against some common attack, but it has little higher computational complexity has only a half payload than DNW. Besides the real-time requirement, some watermarking application has other specific requirements. For the watermarking application of covert communication, the watermark payload is also very important requirement. So a robust algorithm with large data payload and high computational efficiency, the Energy Modulated Watermarking (EMW) algorithm, has been proposed based on modulating the energy of certain image regions, which is suitable for real-time watermarking of JPEG or MPEG streams. During the modulation, a theorem deduced from the Watson’s perceptual model is employed to restrict the modified magnitude of coefficients. Comparing with the Multipath Trellis-coded Modulation (MTM) method, the EMW algorithm has much lower complexity. This system is capable of embedding 2640 bits of information in images with dimensions 704×480 pixels. Experimental results indicate the EMW algorithm is transparent and robust to common processes and attacks. So the EMW algorithm is suitable for high-capacity watermark embedding in the real-time video watermarking applications. How to resist the geometric attacks is one of the most difficulties in the watermarking field. Presently, both the first and second generation watermarking schemes cannot resolve the problem that the video watermarking scheme should not only meet the real-time requirement but also be rubost against the geometric attacks. So a new real-time video watermarking model against geometric attacks has been proposed, which consists of four parts including partly decoding, feature point extraction in compressed domain, watermark embedding in local region and partly recoding. In order to implement the four parts, we gave the mothod to directly extract the robust feature points from compressed video data; and proposed the adjustable-resolution technique for fast decoding and encoding the compressed video frames; finally we proposed a more robust watermarking embedding and extraction method, which embedded the watermark bits repeatly into each local region. From the analysis and experiments, it shows the new model is performed directly in the compressed domain, which can prevent full decompression and recompression and consequently, it largely reduce the computational needs. Moreover, the new model also has the same advantage as the second generation watermarking model.