根据GOP异常进行视频序列剪辑篡改的盲检测

Blind Detection of Video Sequence Montage Based on GOP Abnormality

本文根据对数字视频的剪辑篡改引起MPEG-2编码中GOP异常的现象提出一种盲检测方法.视频序列以一定的GOP结构进行压缩,其运动误差序列在Fourier变换域中存在特定的尖峰,解码后再以不同的GOP结构进行二次压缩,尖峰的分布会发生变化.经不同GOP结构进行多次压缩后,历次处理的痕迹被保留在视频信号中.因此,在视频内容无剧烈变化的情况下,根据尖峰的不一致性可准确定位被替换拼接的视频内容片段.该方法也可根据尖峰所在位置判断视频是否经过二次压缩,并推算第一次压缩的GOP结构,为检测可疑视频内容提供有用信息.实验表明这种基于GOP异常的检测方法能有效检测对视频序列的剪辑篡改.

This paper proposes a blind detection technique to reveal malicious attacks on video contents by video sequence montage.The method is based on an analysis of Fourier transform of the series of motion errors.For an MPEG-2 video sequence with a particular structure of the group of pictures（GOP）,spikes exist in the Fourier transform magnitude of the error series.Multiple MPEG compression with different GOP structures results in variation of spikes in the Fourier domain,leaving clear signatures of the coding history in the video sequence.These spike artifacts are used to detect video splicing,i.e.,replacing certain sections of the video with fake contents taken from other sequences with different GOP structures.This method can also be used to detect suspicious multiple compressions and find the initial GOP structure,providing useful clues to the authentication of video contents.Experiments show effectiveness of the proposed method.