Colluder Detection for Nonlinear Collusion Attacks

We investigate the problem of colluder identification for digital fingerprinting systems under nonlinear collusion attacks. Formulating colluder detection as a binary hypothesis testing problem, we derive the log-likelihood ratio tests for various nonlinear collusion attacks. Utilizing the approximate distribution of the order statistics, we obtain suboptimal detection statistics with low complexity. Compared with the existing correlation-based detectors, these detectors provide substantial improvement in both detection performance and computational complexity.