Robust detection of additive watermarks in transform domains

Most of the watermark detection schemes proposed until now imply the general assumption that the image coefficients, which may carry watermarks, can be perfectly characterised by certain model distributions. However, there are always (small) deviations of the actual coefficient distributions from the idealised theoretical models owing to inherent modelling errors and possible attacks to the watermarking systems. These uncertain deviations, although usually small, may degrade or even upset the performance of the existing optimum detectors that are optimised under idealised assumptions. In this paper, we present a new detection structure for additive watermarking in transform domains based on Huber´s robust hypothesis testing theory. In order to capture the uncertainties, the statistical behaviours of the image subband coefficients are modelled by a contaminated generalised Gaussian distribution (GGD) instead of the perfect GGD. The robust detection structure is derived as a min-max solution of the contamination model and turns out to be a censored version of the optimum probability ratio test. Experimental results on real images confirm the superiority of the proposed detector over the classical optimum detector