结合SIFT校正的视频对象鲁棒水印

Video Object Robust Watermark Combining SIFT Correction

提出一种结合尺度不变特征变换(SIFT)特征点校正和时空域视频对象分割的视频对象水印算法。该算法采用时域和空域结合的方法分割出视频对象,将水印嵌入在视频对象的离散余弦变换域中。在嵌入前提取并注册视频对象的SIFT特征点,在水印检测时,通过与注册机构中特征点集的匹配和计算对视频对象进行几何校正,提取出水印信息。实验结果表明,该算法能够较好地保护视频对象,对几何攻击和常见信号处理攻击具有鲁棒性。

A video object watermark algorithm using Scale-Invariant Feature Transform（SIFT） features and video segmentation based on spatio-temporal information is proposed. The watermark message is embedded in Discrete Cosine Transform（DCT） domain of the detected video object region. Before embedding the watermark, the object SIFT features are detected and registered in advance. In the detecting scheme, the object is corrected by using feature matching, and the message can be detected. Experimental results show that the algorithm can well protect the video object and be robust to geometric attacks and usual signal processing attacks.