Imperceptible Watermark
Definition
A signal embedded in a media file that is statistically detectable by a paired algorithm but falls below the threshold of human perception. Distinct from visible watermarks, which can be removed by simple editing. Imperceptible watermarks survive common transformations such as JPEG compression, resizing, and colour adjustments.
- Detection
- Requires a paired algorithm, not the eye
- Survives
- JPEG compression, resizing, colour adjustment
- Contrast
- Visible watermark, easily cropped out
- Application
- Synthetic media provenance tracking
Common questions
Can an imperceptible watermark be removed by an attacker?+
Determined adversarial processing, such as heavy recompression or format conversion aimed at the watermark, can degrade or remove it, so watermarking is treated as one layer of provenance evidence rather than a guarantee.
Does the absence of a detected watermark prove an image is unaltered?+
No. Absence only shows the paired detector found no signal; many images were never watermarked in the first place, so the watermark's absence is not itself evidence of tampering.
Related terms
- C2PA (Coalition for Content Provenance and Authenticity)
- An open technical standard that embeds cryptographically signed provenance assertions into media files at the point of capture or editing. A C2PA...
- Model Fingerprint
- An unintentional pattern in a generative model's outputs that is characteristic of that model's architecture, training data, or sampling procedure. Unlike watermarks,...
- Regeneration Attack
- A watermark removal technique in which a watermarked image is passed through a diffusion model or variational autoencoder that regenerates semantically similar...
- Robustness-Capacity Tradeoff
- The fundamental tension in watermarking design between the amount of information a watermark can carry (capacity, in bits) and its ability to...
- SynthID
- Google DeepMind's watermarking and detection system for AI-generated content. For images, it modifies pixel values during generation using a trained encoder network....