Video Double Compreesion Detection Based On Prediction Residual

Due to the widely use of social media network and the development of video editing software,digital videos are easy to be accessed in daily life and can be tampered almost costlessly without professional knowledge.It has badly infringed the originality of videos,in which situation it is hard to judge the integrity and authenticity of videos.The degrading general trust and the indispensable validation of video contents have called on the emergent research of video forensics,especially in the justice area.Detection of double compressed videos is one of the most significant issues in video forensics.It is because the tampering process must be conducted on decompression domain and then tampered videos have to undergo the recompression process in most cases.This paper focuses on the GOP not-aligned H.264 double compression detection.The most existing video double compression methods are based on the MPEG-X and can not be applied to H.264.What's more,the only two methods have a bad performance for videos with furious moving objects.Two H.264 double compression detection methods are proposed.The main innovations of this paper are listed as follows:1.A novel double H.264 compression detection scheme based on PRBR analysis is proposed.Firstly,the mask of background regions in each frame is obtained by applying Visual Background Extractor(VIBE).VIBE is an efficient and robust background subtraction algorithm,which can distinguish the background and foreground of each frame at pixel level.Then,the PRBR feature is designed to characterize the statistical distribution of average prediction residual within the background mask.After that,the Jesen-Shannon Divergence is introduced to measure the difference between the PRBR features of the adjacent two frames.Finally,a periodic analysis method is applied to the final feature sequence for double H.264 compression detection and estimation of the first Group Of Pictures(GOP).Eighteen standard testing sequences captured by fixed cameras are used to establish the double compression dataset.Experiments demonstrate the proposed scheme can achieve better performance compared the-state-ofart methods.2.A novel double H.264 compression detection scheme based on LBPPR is proposed.For each frame of a given video,the LBPPR feature is first extracted.The Jensen-Shannon Divergence(JSD)is introduced to measure the difference between the LBPPR features of adjacent two frames.After that,the denoising method is used for the JSD sequence.Finally,a Periodic Analysis method is applied to the final feature sequence to detect double H.264 compression and to estimate the first GOP size.Experiments have demonstrated that the proposed scheme can achieve better performance compared to PRBR methods.The proposed detection schemes have good performance and practicability,which can be applied in video forensic area and chemical video identification.