DWT-Domain Watermark Detection Using Gaussian Mixture Model with Automated Model Selection

This paper presents a discrete wavelet transform (DWT) domain watermark detection approach using Gaussian mixture model (GMM) with automated model selection. More specifically, instead of using traditional expectation maximization (EM) algorithm for parameter estimation in mixture model, where the number of model components need to be fixed in advance, the proposed approach employs the component-wise EM algorithm to realize automatic mixture model selection. And the DWT coefficients with distinct impulse distributional behavior are well characterized. Based on the theory of statistical inference and weak signal detection in nonGaussian noise, a new blind detection algorithm is derived. And the validity of the detector is also tested.