Digital watermarking
Digital watermarking is the practice of embedding an invisible, machine-recoverable signal inside content — image pixels, audio samples, or text — that carries information such as a provenance identifier. A well-designed watermark is imperceptible to people yet survives common transformations like resizing, recompression, and re-uploading.
Watermarking vs. cryptographic signatures
A signature proves, with mathematics, who signed a specific piece of content and that it has not changed. A watermark cannot prove origin on its own — it is a resilience mechanism that helps recover the connection to a signed record when the original metadata is gone. The two work best together: signature for truth, watermark for durability.
How Certivu uses watermarking
Certivu embeds an opaque watermark identifier using frequency-domain techniques — DCT spread-spectrum for images, a DCT frame for WAV audio, and zero-width-character steganography for text and HTML. The identifier is only a lookup handle; it is not proof by itself. The ML-DSA signature remains the source of truth. Certivu does not claim its watermarks are unremovable.
FAQ
What is digital watermarking?
Digital watermarking embeds an invisible, machine-recoverable signal inside content — pixels, audio samples, or text — that carries information such as a provenance identifier. A good watermark is imperceptible to people but survives common transformations like resizing, recompression, and re-uploading.
Is a watermark proof that content is AI-generated?
No. A watermark is a resilience signal, not cryptographic proof. In Certivu, the ML-DSA signature is the source of truth; the watermark only helps recover the link to a signed provenance record after metadata has been stripped.
Can watermarks be removed?
Yes. Watermarks are designed to survive ordinary transformations, not targeted attacks, and Certivu does not claim they are unremovable. That is precisely why the cryptographic signature, not the watermark, is treated as authoritative.