Source device attribution for digital videos

Source device attribution is one of the main tasks of multimedia forensics which aims to identify the device from which a digital image or video originates using blind forensic techniques. Forensics generally rely on unique artifacts created on acquired images or videos during the acquisition process. Photo-Response Non-Uniformity (PRNU) is one of the most important sensor artifacts used in source device attribution due to its unique, random, and robust nature. In this research, we study the effect of video compression on the PRNU noise in video frames and propose new techniques to perform an accurate source device attribution of highly compressed H.264/AVC videos. The proposed scheme for non-stabilized videos was tested on a large set of videos and achieved a high accuracy even on highly compressed videos. We also propose a new scheme for source device attribution of digitally-stabilized video; this scheme was tested on a small set of digitally-stabilized and highly-compressed videos; the results obtained show the effectiveness of the proposed scheme.