Device identification based on H.264 cues

With the rapid development of multimedia techniques, the video can easily be edited and tempered. To determine whether a video has been modified from its original recording has become a critical challenge. Thanks to the flexibility of H.264, we observe that different H.264 video parameters are used by various manufacturers. We propose a video capturing device identification approach by exploring the H.264 cues. Two main video dependent features are exploited: 1) H.264 encoding parameters which rely on motion-vector analysis, entropy coding, encoding quality and level, GOP information, etc. 2) The other is the video encapsulation formats. The former is to judge whether the video is taken by cellphone, while the latter is utilized to identify the video source. In the experimental part, we evaluate our proposed approach on our collected video dataset, and demonstrate the efficiency of our method.