Jamimamul Bakas

Double compression detection is a predominant problem in video forensics. Due to the rapid growth of image/video editing software and multimedia sharing websites, it has become extremely easy to manipulate multimedia data, many times done with malicious intention. One such problem is an intentional modification to videos by carrying out recompression of its (selective) frames. In this paper, we present a forensic solution to detect double compression based forgery in MPEG videos (one of the most commonly used video formats in today's date) as well as to localize the exact region of tampering within the frames. We present a deep learning architecture for the above, which utilizes the video I-frames and the artifacts introduced into those due to double quantization. The proposed method is evaluated using a publicly available standard video dataset to demonstrate the experimental results. Our experimental results prove the efficiency of the proposed technique.