Watermark detection on quantized transform coefficients using product bernoulli distributions

Detection performance of additive spread-spectrum watermarks depends on the statistical host signal model employed to derive the detection statistic. When transform coefficients are heavily quantized, the assumption of a Cauchy or Generalized Gaussian Distribution (GGD) is hard to justify and the estimation of model parameters becomes inaccurate. In this paper we derive a Likelihood-Ratio Test (LRT) based on the product of Bernoulli distributions. The watermark detector is designed to operate on quantized (integer) transform coefficients and therefore permits straightforward integration of the watermarking scheme in popular image and video codecs. Detection performance surpasses the linear correlation detector and is competitive with the computationally more demanding LRT based on a GGD.