Inter-frame Forgery Detection Based on Post-processed MVP with SVM Multi-classifier

As common inter-frame tampering method, frame deletion and duplication are often used to generate falsified videos. In this paper, a video forensic method based on post-processed motion vector pyramid (MVP) is proposed to classify original, frame-deleted and frame-duplicated video. The method is composed of feature extraction and multi-classification. In the first phase, each frame of the video is firstly transformed into grayscale image, and then the motion vector pyramid (MVP) between every two adjacent frames is calculated. Meanwhile, variation factor (VF) is easily calculated by its corresponding motion vector pyramid (MVP). In the second phase, variation factor (VF) is post-processed. Firstly, VFs with normal values and their mean are obtained. Then absolute values of difference between each VF and the mean can be calculated. After that, absolute values are sorted from small to large, and finally distinguishing features are obtained after superposition process. After post-processing, SVM are used as multi-classifier for multi-classification. Experimental results show that the proposed method is more generic. Compared with other algorithms, identification accuracy of the proposed method is higher.