Watermark removal leaves detectable forensic artifacts

A new study introduces Watermark Removal Detection (WRD) as a third evaluation axis for watermark removal methods, alongside attack success rate and perceptual quality. Researchers demonstrate that state-of-the-art removal techniques, while degrading watermark signals without visible distortion, produce distinct statistical artifacts that betray the removal attempt. A classifier trained on these artifacts achieves detection rates at 10^{-3} false positive rate across all tested removal methods. The paper benchmarks leading watermarking schemes against standard removal pipelines and finds no current method balances attack success, perceptual quality, and forensic detectability. The study establishes forensic stealthiness as a necessary requirement for watermark removal.