A generic collusion attack optimization strategy for traditional spread-spectrum and quantization index modulation fingerprinting

Collusion attack is known to be a cost-effective attack against digital fingerprinting. Under the constraint that the fingerprint embedding algorithm and the host signal are unavailable to the adversaries, it is not very easy to perform efficient collusion attack. To solve this problem, we propose a generic collusion attack optimization (GCAO) strategy for traditional spread-spectrum (SS) and quantization index modulation (QIM) fingerprinting. First, we analyze the differential signal of two fingerprinted copies of the same content generated by popular fingerprint embedding algorithms, including both traditional SS-based embedder and QIM-based embedder. Based on the analysis of the differential signal, we construct a fingerprint forensic detector that identifies which embedder is applied and what the embedding parameters are adopted. Then, we improve the earlier proposed SANO collusion attack. Two components outlined above constitute a generic collusion attack optimization strategy. The simulation results show that the proposed forensic detector provides trustworthy performance: the detector can correctly identify the fingerprint embedding algorithm, and the probabilities of estimating the fingerprint length and position are higher than 86 % and 87 % for SS-based embedding and 84 % and 80 % for QIM-based embedding, respectively. The further experimental results illustrate that the proposed GCAO attack provides a better tradeoff between the probability of being undetected and the quality of the attacked copy.