鲁棒视频水印技术研究-智能与分布计算实验室

鲁棒视频水印技术研究

姓名	袁武钢
论文答辩日期	2007.11.07
论文提交日期	2007.11.12
论文级别	博士
中文题名	鲁棒视频水印技术研究
英文题名	Research of Robust Video Watermarking Technology
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;对数幅度调制
英文关键词	Watermarking;Video watermarking;Magnitude modulated watermarking;Log-scaling
中文文摘	视频水印处理技术潜在的应用领域非常广，如数字视频点播系统、DVD版权保护和卫星数字视频传输等等。但目前视频水印技术尚未得到充分的研究，还有很多技术问题没有得到很好的解决，这给研究工作带来了机遇和挑战。对许多视频水印算法来说，一旦失去了含水印视频序列空间上的同步关系，就很难提取水印信息。视频序列作为一系列静止图像的集合，会遭受一些特定种类的攻击(如帧平均、帧去除、帧插入、帧重组、帧速率转换等)，另外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，提出一个鲁棒的视频水印是一个急需要解决的重大难题；如果在压缩视频码流中嵌入水印，很显然应该与视频的压缩编码标准相结合，现有的标准视频编码格式层出不穷，和新兴的视频标准结合的视频水印技术的急需被引入；另外，视频水印算法在应用中经常有实时或接近实时的需求，实时的、鲁棒的、大容量的视频水印将是研究的重点。针对抗几何攻击的数字水印方面，基于特征的局部水印方法，利用ScaleSal检测出的显著区域来同步水印，提出了一种抗几何攻击的局部数字水印算法。此外，还提出了一种抗几何攻击的基于图像内容的水印方案。水印重复嵌入到了基于鲁棒特征点的多个圆形区域，这样就可以抗裁减攻击。每个圆形区域都进行一次PM和二维DFT变换，以此获得RST不变的幅度谱。采用EMW算法修改DFT幅度谱来嵌入每一个水印位。根据所提供的试验结果，可以说该方案在水印的视觉质量影响方面有非常优异的效果。它可以同时抵抗图像裁减和RST攻击。另外，其计算开销相对FMW而言减少较多，从而可以应用于视频水印应用中。在与新兴标准结合，提出与之相结合的视频水印的方面，提出一种基于AVS的数字水印算法(AVS-BW)。通过将两个区域间再嵌入一个水印比特位，实现了两个区域嵌入三个水印比特位，使得水印容量扩大了50%，并且提出了具有自适应性的级差截断索引，增强了水印的鲁棒性。还提出了在AVS中，无需要重新对视频编码，移除高频系数的方法。在文章的最后，给出算法在水印容量、视觉失真和鲁棒性等方面的实验结果。根据这些实验结果，可以得出改进算法在水印容量、视觉失真和鲁棒性三方面上都比DNW算法好。通过对数幅度调制方式，实现了一种具有低复杂性、高鲁棒性及安全性的水印算法。算法通过对系数的幅度值进行对数缩放，使得水印算法的鲁棒性由所有被选系数决定，而不是只受具有较大值的系数的影响，从而提高水印算法的鲁棒性和安全性。在水印容量方面，实现在512×512 图像或视频帧中嵌入512 个比特，基本能满足视频水印应用在水印容量方面的需求。实时性方面，水印嵌入用时为视频正常播放用时的1/6，水印检测用时为正常播放用时的 1/12，完全满足视频水印的实时性需求。
英文文摘	Video Watermarking will be widely used in lots of application areas, such as Video On Demand(VOD), Digital Versatile Disk(DVD) copyright protection, Satellite Digital Video transmission and so on.But the technology of video watermarking hasn't been developed enough.There are still lots of problems to be solved which brings chances and challenges for study. As to many video watermarking algorithms, it's hard to extract watermark once lose the synchronized relations between frames of video. As a series of stationaries pictures set, video sequence may be faced with some specific attacks (such as Frame average, Frame removement, Frame inserting, Frame reconstruct and so on). Besides, to present an rubost realtime watermark algorithm with the ability to resist geometric attacks is the most important problem to be solved. If embeded in the compression video bit streams, the watermark should be compounded with the compression standard. However, the advanced video compression standard emerges one after another incessantly, and lots of new technologies were brought into the standard. An alogrithm based on these new standards needs to be proposed. Finally, large capacity video watermarking is also needed in lots of applications. Based on the invariant regions in the image, a localized image watermarking algorithm against geometrical attacks is proposed. Salient regions detected by Saliency detector are used to synchronize the watermark. How to use the salient regions to construct the invariant regions to attain the translation, scaling and rotation invariance and how to select the embedding regions out of the invariant regions are key problems to be resolved in our paper. The watermark is first geometrically transformed to match the shape of each embedding region and then imperceptibly embedded into corresponding embedding regions. The watermark detection process is blindly done based on the normalized correlation. Another geometrically robust watermarking scheme was proposed based on image local content, which could resist both image cropping and RST attacks and can be used in video because of the low computation quantity. It first extracts the robust feature points, and then partitions the image into multi circular area. Each area is transformed into PM sub-images, then 2-D DFT is performed and the magnitudes are used as the watermark-embedding domain. Finally the watermark is embedded by modifying the DFT magnitudes using energy modulation. Comparing with Fourier-Mellin transform-based watermarking, its computational cost is reduced as only one 2-D DFT is performed. Moreover, the method can overcome the interpolation problem by using PM in the spatial domain because images have a similar scale between neighboring pixels. Based on the Chinese video compression standard AVS, a new watermark algorithm named AVS-BW is proposed. By embedding an additional watermark bit between two regions and using three differential embedding cut-off indexes, the watermark capacity and robustness are successfully improved. And a new method is also proposed in this paper to remove the high frequency components in AVS video sequence. A real-time watermarking scheme with high robustness and security has been proposed based on modulating the log-scaling magnitudes of DCT coefficients, which is most suitable for JPEG images and MPEG streams. The watermark bit is encoded as the sign of the difference between the individual log-scaling magnitude of a group-region and the average one of all group-regions. The log-scaling magnitude can be modulated by modifying the low and middle frequency DCT coefficients imperceptibly. The robustness of scheme is not only dependent on those largest coefficients, but also on the other coefficients with the same proportion. It can embed 512 bits into an image with a size of 512×512, which can satisfy the payload requirement of most video watermarking applications. Moreover, the watermark embedding process only requires one-sixth of the time consumed during normal playing of video, and the watermark detection only requires one-twelfth of that, which can meet the real-time requirements of most video watermarking applications.