Double compression detection for H.264 videos with adaptive GOP structure

As a blind forensic method, double compression detection is valid to multiple manipulations. However, the existing methods only consider to detect the videos with fixed Group of Pictures (GOP). In this paper, we put forward a novel double compression detection method for videos with both fixed and adaptive GOP structure in H.264 videos. Considering that video may contain adaptive GOPs caused by fast moving contents or scene changes, in our double compression detection scheme, temporal segmentation is first used to divide video into static and rapid periods which contain normal fixed and adaptive GOPs respectively. Then, new artifacts based on the sequence of frame’s byte count (FBC) are analyzed. A feature sequence composed of recognizable distances is generated by combining the artifacts in the static and rapid periods of video. Finally, to reveal the intrinsic property of the feature sequence, a scoring strategy is designed to determine whether or not double compression. The experiments demonstrate that the proposed scheme is effective to detect double compression of H.264 videos, and it outperforms other existing state-of-the-art methods.