June 05, 2014, 10:00 am – 11:30 am

Location: CASED, Mornewegstr. 32, S4|14, 5. Stock, Raum 5.3.17, Darmstadt

To prevent such attacks, especially designed codes can be generated and embedded as a watermark message. Even after a collusion attack, it is possible to trace back the participating pirates. Such codes are called collusion-secure fingerprints.
The downside of these codes is that these can be quite large and an embedding as a digital watermark message might not be possible, especially not for very small multimedia. This talk will focuses on the security of the Tardos scheme, the most prominent fingerprinting scheme, for short code lengths.
The two main contributions addressed in this talk are:
1) The asymptotically optimal tracing algorithm by Oosterwijk et al. results to extremely high false positive rates if the parameters are selected as proposed in their paper. We show that this is due to a positive infinite skewness of the scores for innocent users. For more carefully selected parameters, the tracing algorithm shows good results, i.e. low error rates.
2) By neglecting the (unrealistic) assumption that the colluders only use stateless attack strategies, we could derive new attack strategies that result to much higher error rates compared to the known strategies.