Bayesian watermark detection and new perceptual mask based on a spatially weighted Total Variation image prior

In this work we propose a class of bayesian watermark detectors based on a spatially weighted Total Variation (TV) image model. The inherent flexibility of the proposed prior in modelling local image variations, provides us with a novel spatial mask capable to perceptually shape the embedded watermark. We also propose methods to estimate the parameters of the proposed mask, creating watermarks with more energy and in consequence with improved robust properties. Numerical experiments are presented that demonstrate the performance of our proposal with regard to detection sensitivity and the superiority of the mask compared with other existing spatial masking schemes.