Adversarial Removal and Injection of Electric Network Frequency (ENF) Signals in Video Forensics for Cybersecurity

The electric network frequency (ENF) signal embedded in recordings made under grid-powered lighting is widely used to verify recording time and assess authenticity. Its reliability under deliberate anti-forensic manipulation, however, requires further study. This paper demonstrates a signal-processing pipeline that suppresses the original ENF component in a target video and introduces the ENF-bearing luminance component of a donor video recorded at another time. We also examine ENF injection into videos with no usable ENF trace. In eight matched experiments, the ENF re-extracted from the manipulated videos produced normalized cross-correlation values within the range observed for authentic donor videos. Frame-averaged PSNR and SSIM were used to quantify visual fidelity, and two reference-free tests examined artifacts introduced by the pipeline. A batch-boundary phase-discontinuity test was not statistically significant, whereas a notch-shoulder spectral-energy measure separated the original and manipulated videos in this small sample (AUC = 1.00). These results show that ENF substitution can mislead NCC-based timestamp matching but is not necessarily undetectable, supporting the use of complementary forensic checks when ENF evidence is evaluated.