Waterfilling: a connection between information theoretic and perceptually based watermarking

We present a hybrid approach to image watermarking that exploits results from both information theory and perceptual studies, and we extend some of the results of our earlier paper (see Somasundaram, S. and Chandramouli, R., Proc. ISCAS, 2002). We use a waterfilling-type algorithm to establish a connection between information theoretic and perceptually based watermarking. An information theoretic capacity formula for a Gaussian signal over a Gaussian channel is used and some parameters of the corresponding waterfilling-type algorithm are chosen based on a perceptual criterion. Image sub-blocks to be watermarked are chosen based on the local statistics of each sub-block and the embedded watermark strength for each of these sub-blocks is computed based on a perceptual criterion. An iterative algorithm is proposed for this purpose. Experimental results are presented to discuss the trade-off between information theoretic capacity, robustness and perceptual distortion.