Detection of re-compression, transcoding and frame-deletion for digital video authentication

Abstract

In the wake of widespread surfeit of inexpensive and user-friendly digital multimedia alteration software, digital images and videos have lost the unparalleled position they once occupied as `authoritative testament of occurrence of events'. The inherent susceptibility of digital content to malevolent manipulations renders it vulnerable to our skepticism. Establishment of authenticity of digital content is of utmost importance in situations where reliability on fraudulent evidence could have serious consequences. With the intent of tackling a few of the several challenges of the video forensics domain, in this paper we propose a potent DCT coefficient analysis-based forensic technique for reliable detection of re-compression and transcoding in digital videos. This scheme facilitates visually perceptible differentiation between singly-compressed and re-compressed video frames while circumventing the need for undertaking any complicated peak periodicity analysis procedures that are normally associated with traditional DCT-based studies. We also present a unique optical-flow analysis scheme, where, instead of inspecting inconsistencies caused by frame-removal in the entire optical flow sequences of a given video, we focus entirely on the brightness gradient component of this flow. The experiments in this regard substantiate the forensic capabilities of this component and proffer observations conducive to the detection and localization of frame-removal in digital videos. Subjective and quantitative experimentation on a comprehensive dataset under a wide range of experimental set-ups validate the efficacy and resilience of the proposed techniques.