Blind forensics method for GOP period detection in motion compensated video

Abstract

Any tampering in an image or a video is followed by re-compression, thus resulting in double compression. Detecting double compression is a challenge in video forensics. Estimating the period of the Group of Pictures (GOP) in first compression is a clue to detect the double compression, as it gives a way to trace the processing history of the video. This paper, proposes a new method based on the mean absolute motion error sequence for blind GOP period detection in motion compensated video. A cyclo-stationary signal based algorithm, which computes for the largest variability in an n-fold blocked sequence, is adopted for the GOP period extraction. This method gives better results than the existing noise variance method. The proposed method is based on the periodic increase in motion error at the beginning of each GOP, when its motion is estimated from the reference frame in previous GOP. Thus, this method is extended for double encoding detection in a video when GOPs used in both the compressions are not equal, and the GOP period used in first encoding is estimated.