一种INMF的抵抗强几何攻击视频水印构架

A video watermark framework resistant to strong geometric attacks using incremental non-negative matrix factorization

提出一种增量非负矩阵分解(Incremental non-negative matrix factorization,INMF)的抵抗强几何攻击视频水印构架.本构架将加密水印嵌入在宿主视频子块INMF分解的基矩阵若干大系数中,根据时间对比阈值自适应控制水印嵌入强度;利用surf算子提取视频特征点,校正水印检测前的几何参数.对于水印检测端首先进行几何参数校正,并对残余视频中的不完整子块进行预测修复;然后利用完整子块的非负矩阵和系数矩阵恢复完整基矩阵,进而提取完整水印.实验表明,本构架能够有效抵抗高强度剪裁、旋转、平移等几何攻击,能够抵抗帧删除、帧插入等视频特有攻击,与同类方法相比,抵抗几何攻击的能力获得了较大程度提升.

An efficient video watermark framework resistant to strong geometric attacks using INMF （Incremental non-negative matrix factorization） is proposed. Firstly, the encrypted watermark is em- bedded into several big coefficients of the basis matrix obtained by decomposing the host video sub- block by INMF; at the same time, the watermark embedding strength is adaptively adjusted by the time contrast threshold. Secondly, video feature points are extracted by surf operator for the geometrical parameters calibration before watermark detection. Finally, during watermark detection, the geo- metrical parameters are first calibrated, and the incomplete sub-blocks of residual video are restored by prediction. Then, the complete basis matrix is completely recovered by non-negative matrix and coefficient matrix of complete sub-block, and the complete watermark is then extracted. Experimental re- sults show that the framework is able to resist the geometric attacks such as the strong cropping, rota- tion, and translation etc. as well as the video-specific attacks such as frame deletion, and frame insertion. The ability to resist the geometric attacks of this framework is improved greatly over other similar methods.