Temporal Forensics of MPEG Video Using Discrete Wavelet Transform and Support Vector Machine

Abstract

Detection of video forgery is challenging, as finding traces of tampering is a complex task. Digital video forensic is used as an important tool to detect video forgery. Compression history of a video is analyzed to carry out temporal forensics of motion-compensated video such as MPEG-2, MPEG-4, H.263, H.264. Wang and Farid demonstrated that double MPEG compression can detect temporal fingerprints of video forgery, by visually inspecting I-frame prediction error sequence (PES) in the discrete Fourier transform (DFT) domain. This method is prone to human error and can be stressful while analyzing large number of videos. In order to overcome the drawback of this method, here we propose a novel technique to automatically detect video forgery using discrete wavelet transform (DWT) and support vector machine (SVM). A new statistical parameter “γ” related to the difference vector of first-level DWT coarse and detail sub-bands is used for the automatic detection of temporal attacks. Here, we use SVM, which is an effective and efficient classification tool to analyze/recognize data patterns. Experiment is conducted using various SVM kernel functions such as linear, polynomial, quadratic, radial basis function (RBF), and multilayer perceptron (MLP) to classify forged videos. The experimental results demonstrated that the proposed method can efficiently detect video forgery.