Mixing bounded Laplace and Gaussian fingerprints

In a quest to improve the collusion resistance of spread-spectrum multimedia fingerprints with respect to the Gradient Attack, in this paper we realize two facts. One, the expected means of correlation tests performed on collusion attacks that use max-min, median, and averaging filters exhibit different behavior for bounded Laplace and Gaussian fingerprints. Two, by using a balanced mixture of these two distributions to construct multimedia fingerprints, we notice that the most powerful gradient attack vector with respect to the three attack filters can be attenuated substantially, which in turn yields better collusion resistance.