On the effectiveness of MH-based joint-decoders for very short Tardos fingerprinting codes

摘要

Digital content piracy is nowadays really burden-some both for companies and professionals. To overcome this problem, digital data producers usually protect their documents from illegal distributions by leveraging digital fingerprinting techniques for marking each copy of their documents in order to uniquely identify it. Unfortunately, in the case that multiple malicious users jointly collaborate for illegally sharing a document copy, they may forge the embedded fingerprint by simply comparing their copies and modifying the fingerprint's bits that differ in their copies. In [1] an optimal fingerprinting schema has been defined for accusing a guilty user with very high probability under the assumption that the fingerprinting code is sufficiently long. To remove this assumption, several joint-decoding techniques have been recently proposed [2], [3]. In this work, we perform a deep experimental analysis of the Metropolis-Hastings based joint-decoding approach defined in [3], with the aim of assessing the sensitivity to various parameters of the model, thus making it usable in practical contexts.