DCT-domain watermark detector using a normal inverse Gaussian prior

In this paper, a blind statistical method for watermark detection is developed using a symmetrical normal inverse Gaussian (SNIG) prior for modelling the block-DCT coefficients of images. The proposed detector is motivated by the superior ability of the SNIG prior in capturing the statistics of the DCT coefficients as compared to the conventionally used priors. Analytical expressions are derived for the Bayesian log-likelihood ratio and the corresponding mean and variance under null and alternative hypotheses. Extensive experiments are carried out using standard images to study the performance of the proposed detector and the results show that it performs better than several existing detectors in terms of the associated probabilities of false alarm and detection for watermarks of varying strength.