一种INMF的抵抗强几何攻击视频水印构架

摘要

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 geomet -rical 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 coeffi-cient 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 inser-tion .The ability to resist the geometric attacks of this framework is improved greatly over other simi -lar methods .%提出一种增量非负矩阵分解（Incremental non-negative matrix factorization ，INM F）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 INM F 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。