Video Inter-frame Forgery Identification Based on Optical Flow Consistency

qi wang,zhaohong li,zhenzhen zhang,qinglong ma
2013-01-01
Abstract:Identifying inter-frame forgery is a hot topic in video forensics. In this paper, we propose a method based on the assumption that the optical flows are consistent in an original video, while in forgeries the consistency will be destroyed. We first extract optical flow from frames of videos and then calculate the optical flow consistency after normalization and quantization as distinguishing feature to identify inter-frame forgeries. We train the Support Vector Machine to classify original videos and video forgeries with optical flow consistency feature of some sample videos and test the classification accuracy in a large database. Experimental results show that the proposed method is efficient in classifying original videos and forgeries. Furthermore, the proposed method performs also pretty well in classifying frame insertion and frame deletion forgeries. Copyright © 2014 IFSA Publishing, S. L.
What problem does this paper attempt to address?