On optimal collusion strategies for fingerprinting

We study the theoretical performance of linear and non-linear collusion attacks under the assumptions that orthogonal or regular-simplex fingerprints are used, and that the detector performs a linear correlation test in order to decide whether a user of interest is among the colluders. The colluders create a noise-free forgery by applying a mapping / to their individual copies, and then add a noise sequence e to form the actual forgery. They seek the mapping / and the distribution of e that maximize the probability of error of the detector. The performance of mappings such as linear-averaging and interleaving can be compared in this framework. It is also shown that impulsive noise attacks are far more effective than Gaussian attacks.