An Asymmetric Fingerprinting Scheme based on Tardos Codes

Ana Charpentier, Caroline Fontaine, Teddy Furon, Ingemar J. Cox

Introduction

Digital fingerprinting, or traitor tracing, is a critical technique for protecting copyrighted digital content. By embedding unique binary codes into content before distribution, providers can identify users who leak material to unauthorized parties. The Tardos code is widely considered the "gold standard" for this purpose due to its mathematical optimization against collusion attacks, where multiple pirates combine their copies to produce a "clean" version.

However, traditional Tardos codes are symmetric: the Content Provider (CP) knows every user's unique fingerprint. This architecture introduces a significant trust gap, as a malicious or compromised CP could "frame" an innocent Buyer by manually inserting their known fingerprint into a pirated file. While asymmetric protocols—where the Buyer generates their own secret fingerprint—exist, they are difficult to reconcile with Tardos codes. Tardos codes require fingerprints to be drawn from a specific probability distribution based on a secret vector ($p$) known only to the CP. If a Buyer independently selects their bits, the statistical properties ensuring collusion resistance are compromised.

The Decoding-Side Attack

A key contribution of this research is the identification of a "decoding-side" attack. The authors demonstrate that a CP does not need to forge a physical file to frame a Buyer. Since the accusation score calculation depends on the secret vector $p$, a dishonest CP can simply manipulate $p$ during the trial phase. By "cooking the books," the CP can make an innocent Buyer’s score exceed the detection threshold.

Simulation of score evolution for a Tardos code ($m=1000, c=3$). The plot illustrates how manipulating the secret parameters can lead to an innocent user's score crossing the accusation threshold.

Numerical simulations validate this threat. In a scenario with a code length ($m$) of 1000 and a collusion size ($c$) of 3, the authors showed that altering only 20-30% of the secret vector $p$ is sufficient to successfully frame an innocent user.

Proposed Asymmetric Protocol

To address these vulnerabilities, the authors propose a two-phase protocol utilizing commutative encryption and double-blind random selection.

Phase 1: Fingerprint Generation

The CP generates a "deck" of encrypted bits (0s and 1s) following the required Tardos distribution. The Buyer then selects bits to form their fingerprint through a "pick a card" protocol. Commutative encryption ensures that the Buyer cannot see the underlying bits (preventing cheating), while the CP remains unaware of which specific bits the Buyer selected (preventing framing).

Phase 2: Disclosure and Accusation

To maintain practical efficiency, the Buyer eventually reveals a portion of their fingerprint (the "halfword") to the CP. This allows the CP to narrow down suspects without obtaining enough information to forge a complete fingerprint. To prevent the decoding-side attack, all encrypted lists are published to a "Write Once Read Many" (WORM) directory. This public record prevents the CP from retroactively changing the secret vector $p$ to frame a user during the accusation phase.

Key Results and Security Analysis

The proposed scheme is the first asymmetric protocol designed to meet the rigorous statistical requirements of Tardos codes. It is proven to be "totally $c$-secure," meaning it can identify at least one colluder with high probability while maintaining the probability of framing an innocent user below a predefined threshold ($\epsilon$).

While the protocol approximately doubles the required code length ($m \approx 2m_h$) compared to symmetric versions, it preserves the optimal $O(c^2 \log(n/\epsilon))$ scaling. This efficiency ensures the scheme remains viable for large-scale content distribution.

Conclusion and Future Work

The introduction of an asymmetric protocol for Tardos codes significantly enhances the security and fairness of traitor tracing systems. However, limitations remain. The current scheme is restricted to binary Tardos codes and requires a trusted third-party Judge for the final accusation stage. Future research may explore the use of secure multi-party computation to eliminate the need for a trusted judge and extend the protocol to more efficient non-binary codes. Additionally, reducing the cryptographic overhead for implementation on consumer-grade hardware remains a priority.

Summary updated

BibTeX
@inproceedings{charpentier2011asymmetric,
  author = {Ana Charpentier and Caroline Fontaine and Teddy Furon and Ingemar J. Cox},
  title = {{An Asymmetric Fingerprinting Scheme based on Tardos Codes}},
  booktitle = {Information Hiding},
  year = {2011},
  doi = {10.1007/978-3-642-24178-9\_4},
}
Figure