A Novel Deep Learning Framework for Forged Scene Detection in Advanced Video

Abstract

Video forensics is one of the most prominent areas of digital forensics which helps the real world in finding the originality and authenticity of the videos. Existing techniques which focuses on the video forensics relies on the video data stream for the analysis and verification. In recent times there is a new line of research which encountered by monitoring and analyzing the video containers to verify the authenticity of the videos. Furthermore, existing work on the video containers works based on the manual compression and feature extraction which is a time consuming task and require a huge processing power. To address this issue, this paper proposes a novel deep learning framework for detecting the video forgery. The key idea of the proposed deep learning framework is to utilize the dissimilarity between the frames on the video and to extract the meta-information from the video containers. The proposed framework is experimentally tested and validated using high end workstation with Intel Xeon processor and 64 GB RAM and Titan X Pascal GPU card. The experimental validation is carried out on an  average of 100 videos of 360p video format and it is proven that the proposed framework works fine with better accuracy of nearly 98% for the tested samples.