Detection of upscale-crop and splicing for digital video authentication

Abstract The eternal preoccupation with multimedia technology is the precursor of us becoming a civilization replete with astonishing miscellanea of digital audio-visual information. Not so long ago, this digital information (images and videos especially) savored the unique status of ‘definitive proof of occurrence of events’. However, given their susceptibility to malicious modifications, this status is rapidly depreciating. In sensitive areas like intelligence and surveillance, reliance on manipulated visual data could be detrimental. The disparity between the ever-growing importance of digital content and the suspicions regarding their vulnerability to alterations has made it necessary to determine whether or not the contents of a given digital image or video can be considered trustworthy. Digital videos are prone to several kinds of tamper attacks, but on a broad scale these can be categorized as either inter-frame forgeries, where the arrangement of frames in a video is manipulated, or intra-frame forgeries, where the contents of the individual frames are altered. Intra-frame forgeries are simply digital image forgeries performed on individual frames of the video. Upscale-crop and splicing are two intra-frame forgeries, both of which are performed via an image processing operation known as resampling. While the challenge of resampling detection in digital images has remained at the receiving end of much innovation over the past two decades, detection of resampling in digital videos has been regarded with little attention. With the intent of ameliorating this situation, in this paper, we propose a forensic system capable of validating the authenticity of digital videos by establishing if any of its frames or regions of frames have undergone post-production resampling. The system integrates the outcomes of pixel-correlation inspection and noise-inconsistency analysis; the operation of the system as a whole overcomes the limitations usually faced by these individual analyses. The proposed system has been extensively tested on a large dataset consisting of digital videos and images compressed using different codecs at different bit-rates and scaling factors, by varying noise and tampered region sizes. Empirical evidence gathered over this dataset suggests good efficacy of the system in different forensic scenarios.