Collusion resilient fingerprint design by alternating projections

Digital fingerprinting techniques aim to embed unique identification information into digital content distributed to individual users in order to track unauthorized use of multimedia files. Fraudulent users may not only attempt to remove the embedded signatures but also may form coalitions in order to remove the embedded fingerprint and disable tracking. This makes the design of fingerprints challenging. An effective fingerprint should not only carry assigned users information but also guard against the possibility of falsely implicating an innocent user. Furthermore, in possible collusion scenarios, the colluded copies should identify each of the colluders. The embedded fingerprints should be imperceptible to maintain the commercial value of the content and preferably the fingerprint-based identification should survive content preserving signal processing. In this paper we give a precise description of each of these requirements and give a solution framework to obtain a set of fingerprinted images meeting these requirements.