Scaling-based watermarking with universally optimum decoder

In this paper, we propose a blind scaling based image watermarking approach which is highly robust against the noise and gain attacks. Designed independent of the host signal distribution, the proposed scheme can universally be applied in various transform domains with arbitrary distributions. Partitioning the host signal into two patches, we embed the watermark code in one patch while keeping the other one unchanged for blind parameter estimation. We extract the watermark bits using a maximum likelihood decoder approach based on the ratio of the samples summation in each patch. Driving the distribution of the decision variable, we analytically study the performance of the proposed decoder. Employing the ratio of samples as the decision variable makes the proposed scheme invariant to the gain attack. The proposed algorithm is applied to both artificial Gaussian autoregressive signals as well as various test images. The robustness of the proposed decoder to various host signal distribution is verified. Experimental results confirm the validity of our model for low frequency components of natural images and its high robustness against common attacks.