Quantcast

Forensics and counter anti-forensics of video inter-frame forgery

Research paper by Xiangui Kang, Jingxian Liu, Hongmei Liu, Z. Jane Wang

Indexed on: 12 Jul '15Published on: 12 Jul '15Published in: Multimedia Tools and Applications



Abstract

Among different types of video manipulations, video inter-frame forgery is a powerful and common tampering operation. Several forensic and anti-forensic techniques have been proposed to deal with this challenge. In this paper, we first improve an existing video frame deletion detection algorithm. The improvement is attributed to the combination of two properties resulted from video frame deletion, the periodicity and the magnitude of the fingerprint in the P-frame prediction error. We then analyze a typical anti-forensic method of video frame deletion, and prove that the fingerprint of frame deletion still can be discovered after being anti-forensically modified. We thus further propose a counter anti-forensics approach by estimating the true prediction error and comparing it with the prediction error stored in videos. We show that the detection algorithm is not only useful in detecting video frame deletion, but also useful for detecting video frame insertion. Compared with the existing counter anti-forensics, our proposed approach is robust when different motion estimation algorithms are used in the initial compression. Furthermore, the forensics and counter anti-forensics are combined to perform a two-phase test to detect video inter-frame forgery. A Video Inter-frame Forgery (VIF) game, which is zero-sum, simultaneous-move, is defined to analyze the interplay between the forger and the investigator. Mixed strategy Nash equilibrium is introduced to solve the VIF game and we can obtain the optimal strategies for both players. Experimental results show that the proposed forensic and counter anti-forensic methods not only outperform existing methods in detecting frame deletion and anti-forensics, but also outperform them in the VIF game.