Frame Rate, Not the Grid: A Confound in Single-Clip Optical ENF Video Forensics

I. INTRODUCTION

• The optical ENF is a rolling-shutter imprint of mains frequency, whose regional drift has supported timestamp verification and recording-region identification.A single-clip nominal-grid reading is proposed as a way to identify the grid without a reference database and potentially distinguish generated footage.Source: §I · p. 1
• Prior work modeled frame-rate aliasing as an extraction nuisance; this letter examines its effect on single-clip grid identification and video authenticity judgments.The paper surveys 1,072 clips, tests the mechanism with controlled synthesis, and introduces a comb-aware mitigation.Source: §I · p. 1
• The paper establishes that frame-rate aliasing confounds single-clip grid identification and can undermine region identification and genuine-versus-generated discrimination.It distinguishes the known aliasing phenomenon from its forensic consequences and proposes a mitigation.Source: §I · p. 1

II. THE FRAME-RATE CONFOUND

• Sequential row exposure resolves 100 Hz or 120 Hz mains-driven flicker as spatial banding, so estimation searches within ±2 Hz of those frequencies.The row rate H/Tro is much higher than the frame rate.Source: §II · p. 1
• A per-frame brightness or codec variation forms a comb at integer multiples of fps, and an in-band tooth is indistinguishable from genuine flicker.The confound condition is met when k·fps lies within ±2 Hz of the flicker line at 2·fgrid.Source: §II · p. 2
• At common rates, frame-rate harmonics fall in the flicker band: 25 fps yields a 50 Hz reading, while 24 and 30 fps yield 60 Hz.The recovered grid follows frame rate rather than electricity supply; it is reliable only when no frame-rate tooth falls in either measurement band.Source: §II · p. 2, Table I · p. 2

A. Data

• The survey assembled 1,072 clips from six text-to-video systems and multiple genuine-video sources covering both 50 Hz and 60 Hz supplies.CogVideo renders at 16 fps and is excluded from grid analysis; clips were analyzed over segments of up to 120 seconds.Source: A. Data · p. 2
• The study records flicker-line prominence as a genuine-versus-generated feature and the recovered nominal grid for each clip.The focused mechanism analysis covers 362 clips across thirteen sources.Source: A. Data · p. 2
• 0.509 AUC (95% confidence interval 0.469-0.549) shows flicker prominence does not separate genuine from generated video on 804 measurable clips.The value is indistinguishable from chance; generated video carries an in-band line at real-comparable strength.Source: B. The flicker feature does not… · p. 2

C. The recovered grid is fixed by the frame rate

• Every clip at 24, 23.98, 29.97, or 30 fps returned 60 Hz, while every clip at 25 or 50 fps returned 50 Hz, across 346 clips.There was no exception; non-degenerate-rate clips broke the one-grid-per-rate mapping, including a real 60 Hz recording at 28 fps.Source: C. The recovered grid is fixed by the… · p. 2
• A confident in-band peak can be a frame-rate comb tooth rather than a grid line, despite its high absolute energy.For 24 and 30 fps generators, the peak's leave-one-out excess was 1.05, below the 95th-percentile null of 3.25.Source: C. The recovered grid is fixed by the… · p. 2
• At 24 fps, the synthetic no-grid spectrum has a 120 Hz comb tooth and an empty 100 Hz band; a genuine 50 Hz grid at 27 fps produces a single 100 Hz line.The 27 fps example has no neighboring comb tooth beside the genuine line.Source: Figure 1 · p. 3

D. Controlled synthesis

• With no genuine grid, the naive estimator returned a confident frame-rate-predicted grid at every degenerate rate and never abstained.It returned 60 Hz at 24 and 30 fps, and 50 Hz at 25 and 50 fps.Source: D. Controlled synthesis · p. 3, Figure 2 · p. 3
• With a genuine grid present, the estimator read it correctly at every frame rate on the majority vote.The true grid was held fixed while only frame rate varied, across five random seeds per rate.Source: D. Controlled synthesis · p. 3, Figure 2 · p. 3
• The controlled results attribute the confident wrong answer to the frame-rate comb when no grid is present, rather than to a weak genuine signal.With no grid, the naive read matches the frame-rate prediction at degenerate rates and wanders elsewhere.Source: D. Controlled synthesis · p. 3, Figure 2 · p. 3

E. A temporal cross-check

• Generator comb teeth are, on average, more phase-coherent than genuine recordings captured at non-degenerate rates.The comb tooth is locked to the camera clock, while a genuine grid line wanders with the mains.Source: E. A temporal cross-check · p. 3
• A genuine recording at a degenerate rate can also be indistinguishable on phase coherence because its flicker coincides with a comb tooth.The passage presents coherence as an estimator-independent hint, not a decision on its own.Source: E. A temporal cross-check · p. 3
• The temporal comparison is between generator teeth and genuine recordings captured at non-degenerate rates, not all genuine recordings.At degenerate rates, genuine flicker coincides with a comb tooth.Source: E. A temporal cross-check · p. 3

IV. A COMB-AWARE MITIGATION

• The detector spans the second, fourth, and sixth flicker harmonics so a genuine grid can be corroborated across harmonics not populated by a single frame-rate tooth.The approach also supplies a relative prominence ratio and adds a dedicated comb class to training.Source: §IV · p. 3
• On 40 synthetic combs with no grid line, the deployed detector labeled all 40 genuine, while the comb-aware detector labeled all 40 comb, with no genuine clip misclassified as a comb.This is a controlled-condition result.Source: §IV · pp. 3–4
• Separating a faint genuine line from a comb on a single real clip still requires an external reference.The demonstrated rejection result is explicitly under controlled conditions.Source: §IV · p. 4

• The recovered grid follows the camera frame rate rather than the mains supply, and the estimator can return a confident wrong answer instead of abstaining.Source: §V · p. 4
• A 100 or 120 Hz line is not evidence of authenticity, and a short-clip grid reading is not assessable unless the frame rate is non-degenerate with both grid standards.Source: §V · p. 4
• Forensic reports should pair a single-clip grid reading with the frame rate or withhold it at degenerate rates.Source: §V · p. 4
• A comb-aware estimator restores the distinction under controlled conditions and recovers the true grid at non-degenerate rates.Source: §V · p. 4
• Separating a faint genuine line from a comb in degraded single clips without a reference remains future work.Source: §V · p. 4