Exposing Video Forgeries by Detecting Misaligned Double Compression

Abstract

Powerful and fast video editing software tools have made video tampering an easy work. In video forgeries, re-compression is one of the inevitable post-processing, so it is usually explored in video forensics works. However, most works only consider aligned double compression---the order of frames remains unchanged before and after re-compression. In some cases, misaligned double compression could happen as well in video forgeries, such as video splicing, etc. In this paper, we propose an effective method to detect misaligned double compression. Based on extensive experiments and analysis, we found that H.264/AVC compression generates different strengths of blurring artifacts in different types of frames. After misaligned re-compression, such blurring artifacts introduced by the previous compression would be preserved in the re-compressed frames. Based on this observation, we propose a compact yet effective feature vector (MBAS) to expose video forgeries. The experiments evaluated on standard test sequences with a variety of encoding parameters have shown the effectiveness of the proposed method.