Inter-Frame Forgery Video Identification Based On PPoMVP

With the rapid development of society, science and technology,the popularity of digital devices is growing increasingly. Meanwhile, simple and powerful video editing softwares with vivid tampering effect can be found everywhere.On one hand, these video editing softwares enrich people’s lives, they lead to frequent video tampering event on the other hand. Because digital videos are one of the seven types of judicial evidence, identifying the authenticity of digital videos will provide a strong technical support for social fairness and justice. Therefore, the study of passive video forensics is of important practical and legal significance.In real life, intentional inter-frame falsified video is generally constructed by inserting or deleting a complete event. Compared with image forensics, video forensics started relatively late, therefore, there are a few algorithms that identify this kind of meaningful inter-frame falsified videos at present. In this paper, an effective algorithm based on PPoMVP (post-processing of motion vector pyramid) is proposed to detect inter-frame forgeries which are difficult for eyes to observe. Firstly, the MVP feature of the given video is extracted, then PPoMVP feature can be obtained by optimizing the MVP feature, and finally SVM (support vector machine) is used to distinguish inter-frame forgeries from original videos, experimental results show the effectiveness of this method. The main research work and innovations are as follows:1. The detection algorithm for inter-frame falsified videos based on MVP is studied, and it is compared with other existing classical algorithms. The MVP feature of original video is approximately continuous, while in forgeries the consistency will be destroyed. Therefore, MVP can be used as an effective feature to detect inter-frame falsified videos. Experimental results show that the proposed algorithm can effectively distinguish inter-frame forgeries from original videos, and it is more generic than algorithms that based on optical flow or velocity field.2. A new algorithm for identifying inter-frame video forgeries based on PPoMVP is proposed. Video tampering will not only change the MVP of tamper point significantly, but also leave a slight trace on other normal points. Through a series of post-processing of MVP features, i.e., mean removal, accumulation, calculating histogram, the irrelevant information can be suppressed effectively, and the traces of video tampering can be fully utilized. Finally, the PPoMVP features of 80 percent video samples are are used to train the support vector machine, and the left samples are used to test the classification accuracy of the proposed algorithm. The experimental results show that the post-processed MVP feature greatly improves accuracy of identifying inter-frame video forgeries.