许多数字图像和视频水印方案都无法抵挡几何变换攻击,即使是轻微的旋转、缩放、平移也将会破坏水印的同步,这就使得水印难以检测出来。为解决此难题,提出了一种新颖的视频水印算法。该算法利用了运动矢量特征的相对关系不易随几何失真变换而改变的特性,使得水印具有内在的抗几何失真的鲁棒性。该算法中,水印嵌入和检测时,不仅可根据平均运动矢量的方向和幅度确定水印模板的位置和方向,而且水印检测不需要原始视频。实验表明,该水印算法对几何变换攻击具有较强的鲁棒性。
For many digital image and video watermarks, geometric distortions are one of the weaknesses. Even slight rotation, scaling and translation can destroy the synchronization and invalidate the detection of blind watermarks. In this paper, we present a novel video watermarking scheme that inserts a watermark intrinsically resistant to these distortions by exploiting the characteristics of the motion vector. We find that the relative relationship between characteristics of the motion vector cannot be altered with the geometric distortions easily. So the position and direction of watermark are determined according to the average values o( the magnitude and direction o(motion vectors. In addition, another advantage of the novel watermarking scheme is that the original video is not needed in watermark detection process. Experimental results show that the scheme is rather robust to the common geometric distortions including rotation, scaling and cutting.