Multimedia Forensics and Security Using Electric Network Frequency (ENF) Signal Analysis and Information Processing

Ericmoore T Ngharamike

doi:10.25907/00992

The exponential growth of digital multimedia technologies has transformed the modes of communication, information dissemination, and evidence acquisition across society. Video and audio recordings have become critical tools in journalism, entertainment, national security, and legal proceedings, owing to their richness in contextual and temporal information. However, the same technological advancements that enable their widespread creation and sharing have also increased their vulnerability to sophisticated tampering and anti-forensic attacks. The emergence of powerful generative models, such as generative adversarial networks (GANs), coupled with accessible editing tools poses significant threats to the authenticity and integrity of multimedia evidence, undermining public trust and justice administration. Electric Network Frequency (ENF) analysis has emerged as a promising forensic technique for validating the authenticity, integrity, time, and location of multimedia recordings. ENF signals originate from minor fluctuations in power grid frequencies (50 Hz in Australia and Europe, 60 Hz in North America) due to variations in supply and demand. These fluctuations create unique temporal signatures that are passively embedded in audio recordings made near power sources and in video recordings captured under mains-powered illumination. Despite its widespread adoption as a reliable environmental watermark for forensic verification, the robustness of ENF-based techniques against adversarial manipulation remains underexplored. This thesis will address critical knowledge gaps by systematically investigating the security, vulnerabilities, and forensic utility of ENF signals in digital multimedia recordings. The major contributions will include: • An extensive survey of ENF-based multimedia forensics, detailing current applications, methodological advancements, and emerging challenges. • A comprehensive analysis of the impact of light sources on ENF capturing and extraction from video recordings, demonstrating how lighting conditions, bulb types, and illumination dynamics affect ENF signal fidelity. • A novel superpixel-based ENF extraction technique for intra-grid location estimation of smartphone video recordings, enabling accurate regional source attribution within power grid networks. • An investigation into rolling shutter readout times for ENF-based camera identification, exploiting camera sensor characteristics to enhance device attribution accuracy. • The development of an ENF manipulation framework demonstrating the feasibility of ENF signal removal and replacement in video recordings, thus revealing potential vulnerabilities of ENF-based verification to anti-forensic attacks. This thesis systematically investigates the security and forensic applicability of ENF-based techniques through multiple interconnected studies. First, it presents a comprehensive survey and framework of ENF-based multimedia forensic applications, mapping existing methodologies, challenges, and future research directions. It then proposes an improved ENF extraction method using superpixel segmentation to enhance location estimation accuracy for smartphone videos within interconnected regions in a power grid. The thesis further demonstrates the feasibility of camera identification based on rolling shutter readout time combined with ENF characteristics, enabling forensic analysts to attribute video recordings to specific recording devices. Additionally, it analyses the impact of different lighting conditions and bulb types on ENF capture in videos, providing practical insights for optimal ENF extraction. Critically, the thesis demonstrates the feasibility of ENF forgery by developing a systematic framework to remove original ENF traces from a video and replace them with ENF signals extracted from a donor recording. Experimental results show that such manipulation can produce forged videos with high temporal ENF correlation to the donor signal while maintaining visual fidelity, thereby exposing the vulnerability of ENF-based authentication to sophisticated anti-forensic attacks. The research contributes significantly to the fields of multimedia forensics and security by advancing ENF extraction methodologies, exposing critical forensic vulnerabilities, and guiding the development of more resilient authentication frameworks against future adversarial threats.

Multimedia Forensics and Security Using Electric Network Frequency (ENF) Signal Analysis and Information Processing

Files and links (1)

Abstract

Details

Metrics