| Nowadays，people are surrounding with numerous digital multimedia informationwhich makes our lives colorful but confuse the truth. The more widely multifunctionalediting software is used, the worse confidence crisis is for digital video. Therefore, theresearch on authenticity of digital video forensics is urgent and difficult in digitalforensic. Exposing the double-compression can provide important clues for videoauthenticity forensic. MPEG-2as the main kind of digital video format, its research ondouble-compression tamper is significant.First, this paper proposes a method to estimate the MPEG-2’s GOP structuresblindly to solve this precondition. This method utilizes the different type frames’ DCTcoefficient energy without accessing MPEG-2bit-stream. Analyzing the extremumsequence diagram to estimate the GOP structure(size of GOP N and the D-value ofadjacent P-frames M). Experiment results show this method is more accurate than fastnoise estimation for N, even can estimate M for resolution greater than CIF format.Secondly, this paper detects I-frames’ double-compression by the improvedmethods of JPEG double-compression test, even to estimate the length of priorquantization step. After extracting the I-frames from the MPEG-2sequences, weanalyzing the I-frames’ double-compression with the methods combined the DCTHistogram and its Fourier Spectrum Spikes’ number. We also estimate the priorquantization step of I-frames accurately based on it.The estimation accuracy of priorquantization step of I-frames is above90%.Finally, this paper detects MPEG-2double-compression systematically. We designa Sliding Window Hopping for Extremum (SWHE) detection algorithm to analyze theperiodic deranged extreme value in GOP level; some improved scheme for P-frames’motion error and content-changed features are also analyzed for non-GOP level.In conclusion, this paper learns the current GOP structure as the prerequisite, thendivides the double-compression detection as two major categories to design the systemimplementation. The research results are beneficial to enrich and improve the theoryand technology in the field of digital video forensics. |
