Estimating Video Fingerprints

S. He and D. Kirovski (USA)


Video watermarking, fingerprinting, cryptanalysis, signal estimation.


In this paper we propose a simple signal processing proce dure that aims at removing low-frequency fingerprints em bedded in video signals. Although we construct an instance of the attack and show its efficacy using a specific video fingerprinting algorithm, the generic form of the attack can be applied to an arbitrary video marking scheme. The pro posed attack uses two estimates: one of the embedded fin gerprint and another of the original content, to create the attack vector. This vector is amplified and subtracted from the fingerprinted video sequence to create the attack copy. The amplification factor is maximized so to achieve a de sired level of visual fidelity. In the conducted experiments, the attack procedure on the average halved the expected detector correlation compared to additive white gaussian noise. It also substantially increased the probability of a false positive under attack for the addressed fingerprinting algorithm.

Important Links:

Go Back