Tardos fingerprinting codes in the combined digit model

We introduce a new attack model for collusion-secure codes, called the combined digit model, which represents signal processing attacks against the underlying watermarking level better than existing models. In this paper, we analyze the performance of two variants of the Tardos code and show that both variants can accommodate the new model and resist collusion attacks with only a modest increase in code length as compared to the results for the commonly used restricted digit model.