Stefan Katzenbeisser

We formalize a new attack model for collusion secure codes, incorporating attacks on the underlying watermarking scheme as well as cut-and-paste attacks traditionally considered for collusion secure codes. We use this model to analyze the collusion resistance of two versions of the Tardos code, both for binary and nonbinary alphabets. The model allows us to consider different signal processing attacks on the content, namely the addition of noise and averaging attacks. The latter may result in content segments that have multiple watermarks embedded. We study two versions of the q-ary Tardos code in which the accusation method has been modified so as to allow for the detection of multiple symbols in the same content segment. We show that both variants yield efficient codes in the new model, parametrized for realistic attacker strengths.

In a forensic watermarking architecture, a buyer-seller protocol protects the watermark secrets from the buyer and prevents false infringement accusations by the seller. Existing protocols encrypt the watermark and the content with a homomorphic public-key cipher and perform embedding under encryption. When used for multimedia data, these protocols create a large computation and bandwidth overhead. In this correspondence, we show that the same functionality can be achieved efficiently using recently proposed secure watermark embedding algorithms.

We propose a novel method for embedding robust forensic tracking watermarks to complement encryption in electronic music distribution applications. The watermark is embedded at the player during the AAC decoding process through modification of scale-factors in pre-defined frequency bands. It thereby modulates the short-time envelope of the decoded audio in the corresponding band. The resulting watermark is robust to various attacks such as re-compression and acoustic transmission. The method, due to its negligible computational overhead, can be used in resource constrained devices such as portable players.