Video Forgery Detection Based On Passive Authentication

With the advent of multimedia era, digital videos have integrated into people’s daily life through the Internet and smart phones. In the meantime, they have gradually become a vital part of judicial evidence. However, the advanced technology and sophisticated software have rendered digital im-ages and videos exposed to forgery. The authentication of multimedia con-tent is of significance. In this paper, we propose four passive authentication algorithms of double compression detection and content-based inter-frame forgery detection.We propose three double compression detection algorithms by ex-tracting statistical features to expose double compression artifacts and adopt-ing machine learning technique. First, a double compression detection algorithm based on first digit distribution is proposed. The first digit dis-tribution of quantized AC (Alternating Current) coefficients is extracted and fitted with logarithmic law. Furthermore, a serial Support Vector Ma-chine architecture is devised to estimate the original bit rate of double-compressed video. Second, we propose a double compression detection algorithm based on Markov statistics. Though mathematical analysis of MPEG-4quantization methodology, the characteristics of doubly quan-tized DCT (Discrete Cosine Transform) coefficients are discovered. The DCT difference array is modeled by Markov random process to detect dou- ble compression. Third, a primary quantization parameter estimation al-gorithm is proposed. The probability distributions of DCT coefficients, first-order and second-order difference arrays are extracted and then clas-sified by multi-class classifiers to estimate the quantization parameter of double-compressed video.For content-based inter-frame forgery detection, we propose an inter-frame forgery type identification algorithm based on optical flow and anoma-ly detection. The footstone is that optical flow in original video changes continuously, while forgery will introduce discontinuity points in optical flow sequence. We propose to calculate optical flow variation factor and adopt anomaly detection to locate discontinuity points. The inter-frame forgery type is then identified by analyzing the characteristics of disconti-nuity points and optical flow sequence.