Near Optimal Detection of DCT-Domain Watermarks using Alpha-Stable Models

Abstract

Digital watermarking can be formulated as a communication task. And the performance of a watermarking scheme relies heavily on the design of the detector. This paper proposes a novel near optimal watermark detector for still images in the discrete cosine transform (DCT) domain. First, the alternative current (AC) DCT coefficients are statistically modeled by symmetric alpha-stable distributions. Then, by resorting to well grounded theoretical results of statistical detection theory, the detector structure in the additive alpha-stable noise interference is derived. Through the approximation to the locally optimum score function, the near optimal detector can be easily implemented. The experimental results demonstrate the superiority of the new detector to correlator detector and generalized Gaussian distribution (GGD) based detector.