Privacy-Preserving Probability Fingerprinting Scheme with Collusion Resistance for Relational Databases

There are numerous fingerprinting schemes with privacy protection that are capable of simultaneously ensuring data privacy and providing liability guarantees, they still face challenges such as low data utility and poor fingerprint robustness. In this work, we propose the Privacy-Preserving Probability Fingerprinting Scheme with Collusion Resistance (P3FP), which achieves privacy and liability guarantees in a one-step manner by utilizing random response noise satisfying differential privacy (DP) as fingerprint embedding into the database. With P3FP, we initially integrate the Tardos code to defend against collusion attack and enhance fingerprint robustness. Then, we leverage the conditional correlation between data points to improve the probability assignment of random response, enhancing the data utility of the fingerprinted database. Experiments are evaluated on two real-world datasets. The results show that under the same privacy budget, P3FP outperforms the state-of-the-art fingerprinting schemes in terms of data utility on large-domain datasets and is capable of resisting random flip and collusion attacks. Even if 50% of the data is deleted, tampered with or updated, the fingerprint information can still be recovered.