Content authentication and identification under informed attacks

We consider the problem of content identification and authentication based on digital content fingerprinting. Contrary to existing work in which the performance of these systems under blind attacks is analysed, we investigate the information-theoretic performance under informed attacks. In the case of binary content fingerprinting, in a blind attack, a probe is produced at random independently from the fingerprints of the original contents. Contrarily, informed attacks assume that the attacker might have some information about the original content and is thus able to produce a counterfeit probe that is related to an authentic fingerprint corresponding to an original item, thus leading to an increased probability of false acceptance. We demonstrate the impact of the ability of an attacker to create counterfeit items whose fingerprints are related to fingerprints of authentic items, and consider the influence of the length of the fingerprint on the performance of finite-length systems. Finally, the information-theoretic achieveble rate of content identification systems sustaining informed attacks is derived under asymptotic assumptions about the fingerprint length.