Robust video source recognition in presence of motion stabilization

Abstract

Video source attribution is getting a growing interest from researchers, law enforcement officers and forensic analysts. The capability of linking a video recording with its source device enables to search out who has generated a video recording. Such a feature finds immediate application in fighting against technology enabled crimes such as digital piracy and child abuse online. Currently, the most powerful techniques rely on the unique noise traces left by each camera sensor within any visual content, widely known as Photo Response NonUniformity. However, in the case of videos, the increasing adoption of digital motion stabilization interferes with the extraction of reliable noise patterns. In such a context, this paper describes a novel methodology for creating a robust reference video PRNU from still images for source camera recognition. Moreover, we provide a novel optimized strategy to compare two different PRNUs extracted from videos in presence of motion stabilization. The conducted experimental evaluation highlights the strength of the proposed methods.