基于NMF和SURF的视频帧间复制粘贴伪造盲检测

摘要

针对视频帧间复制粘贴伪造,本文提出一种基于非负矩阵分解(Nonnegative matrix factorization,NMF)和加速稳健特征(Speed-up robust features,SURF)的帧间复制粘贴伪造盲检测算法.通过对视频帧进行小波变换,提取低频系数矩阵进行非负矩阵分解,将得到的系数矩阵作为视频帧的特征表示衡量帧间的相似性,根据相似度变化趋势判断视频帧间的连续性,从而确定疑似伪造复制粘贴序列的首帧及尾帧,并通过SURF特征匹配进行二次判定.实验结果表明,本文所提出帧间伪造检测算法对连续多帧的复制粘贴伪造具有较好的检测效果,避免了逐帧比对,降低了时间复杂度.%Interframe and intraframe copy-paste forgery are two typical video forgery means.Interframe manipulation is easily done with the video editing software.An algorithm based on nonnegative matrix factorization(NMF) and speed-up robust featuare(SURF) is proposed for detecting interframe copy-paste forgery.Frames are transformed via wavelet.And low-frequency band is further decomposed by NMF.The NMF coefficient matrix is served as the video frame feature descriptor.The variation of frame similarity is used for locating the forged first and last frames via measuring frame similarity.The starting and ending frames mark the suspicious frame sequence,and these frames are further investigated by SURF.The presented approach avoids frame-by-frame match and decreases the time complexity to O(n).The experiment demostrates that the proposed algorithm performs well for interframe video copy-paste forgery detection.