Optimum watermark design by vector space projections

We introduce an optimum watermark embedding technique that satisfies common watermarking requirements such as visual fidelity, sufficient embedding rate, robustness against noise and tolerance to benign signal processing, while optimizing one of these requirements. The algorithm distinguishes itself from other watermark optimization techniques in its flexibility for incorporating constraints and in assuring convergence to the globally optimum point when all constraints are convex. The proposed scheme is a natural extension of set-theoretic watermark design and inherits all the advantages of it. A watermarked image is first obtained by POCS, then optimum watermarked image is determined by an iterative search procedure based on a simple bi-section method. Experimental results are presented to illustrate the effectiveness of the method.