Zero-knowledge watermark detection resistant to ambiguity attacks

A zero-knowledge watermark detector allows an owner to prove to a verifier that an image in question indeed contains the owner's watermark without revealing much information about the actual watermark. In such a scenario, the owner publishes a committed watermark before watermark detection so as to show that she knows the watermark before the detection. However, this does not imply that the owner can prove that she knows the watermark before the work appeared in the public. One well known counter example is the invertibility/ambiguity attacks where an adversary can create an ambiguous situation by deriving a forged watermark from a published work, and commits the forged watermark. Furthermore, the adversary may derive a watermark from existing non-watermarked images in the public domain and later claim ownership of them. One solution is to enforce certain constraints on the valid watermarks. In this paper we propose a zero-knowledge watermark detector that prevents the owner from cheating by ambiguity attacks. In addition, it allows the owner to publish a large number of works with different watermarks, while committing only one secret.