Résumé: The video forensics capabilities are constantly improving in terms of evidence accumulating, analysis, processing, and storage. Video forensic analysis involves scientific investigation, comparison, and/or assessment of video files that are considered as proof in the court. In this paper, we focus on inter-frame video forgery detection and localization with respect to frame inserting and deleting that are essential from digital video forensics perspective. We design a 3-dimensional convolutional neural network (3DCNN) model for detecting video inter-frame forgery and localize forgery using multi-scale structural similarity index measurement algorithm. We introduce an absolute difference algorithm to differentiate video frames from each other which minimize the temporal redundancy and identify forgery artefacts within the video frames. This improves the 3DCNN efficiency and accuracy in detecting frame insertion and deletion forgery. The proposed model outperforms existing models in terms of accuracy, precision, recall, and F1 score in various post-processing operations, compression rates, and video length.

Langue: Anglais