针对现有大部分抗几何攻击的水印技术不具有实时嵌入的特性，提出了一种新颖的基于DCT域特征块的抗几何攻击的视频水印方案。该算法在DCT域中通过选取一定的特征块作为水印嵌入点，再利用由特征块计算得到的同步块来在水印检测前进行几何校准，即在检测端首先定位特征块，再由此计算得到同步块，并对同步块采用邻域搜索方法精确估计视频帧的几何失真。视频帧校准后，系统再次通过提取特征块来进行水印检测。实验表明，该水印方案能够有效地抵抗旋转、平移、缩放等基本几何攻击，并具有较好的实时性和安全性。

A new geometric invariant video watermarking method is proposed based on DCT feature blocks that can satisfy the real-time embedding requirement. In the algorithm, certain number of DCT feature blocks are selected as the embedding points. The synchronization block is calculated accordingly for frame registration before detection is conducted. On the detection side, the feature blocks are firstly extracted for the localization of the synchronization block. And then neighborhood search is implemented on the synchronization block to accurately estimate the geometric distortion. After the frame registration, feature blocks are extracted again for the final watermark detection. Experimental results show that the watermarking system can resist to geometric attacks well with satisfying good invisibility, real-time and security properties.