A dual watermarking and fingerprinting system

We propose a public-key watermarking system, where as usual, all the copies of a protected object are identically marked, but where each user has a distinct secret detection key. All such detection keys are different from the secret embedding key. Even by gaining the knowledge of a relatively large number of detection keys, an adversary cannot remove the secret marks from the protected content. We show that an attacker who has access to one detection key can always fool the corresponding watermark (WM) detector, but not other WM detectors. Also, in that process, the attacker necessarily inserts a fingerprint (FP) in the modified content. Our WM/FP system significantly improves upon the best known asymptotic resistance to collusion attacks to O(N log N) from O(N¹4/).