Asymptotically Optimal Detection of DCT-Domain Watermarks Using an Approximate Alpha-Stable Prior

This paper proposes an asymptoticallyoptimal watermark detector in Discrete cosine transform(DCT) domain images. The Gaussian mixture model asan analytical approximation of alpha-stable distribution ischosen to statistically model the Alternative current (AC)DCT coeffcients. Given the additive embedding modeland based on the theory of weak signal detection in nongaussian noise, the derivation of the detector is presented.The experimental results demonstrate the superiority ofthe new detector over correlation based detector and Generalized Gaussian distribution (GGD) based detector.