Determining the Capacity Parameters in PEE-Based Reversible Image Watermarking

In the existing prediction-error expansion (PEE)-based reversible image watermarking schemes, the capacity parameters are determined in a recursive manner until the payload is just accommodated. This class of methods requires many rounds of embedding iterations, especially when the payload is high, and therefore, is computationally inefficient. Moreover, when multiple capacity parameters need to be determined, the previous methods cannot guarantee optimality in the capacity-distortion sense. In this work, a capacity-distortion optimization (CDO) framework is built to estimate the optimal capacity parameters. We prove that the CDO problem is convex for any embedding rates, permitting efficient solution. The estimated capacity parameters then serve as starting point to facilitate a local search algorithm to find the optimal capacity parameters with much less rounds of embedding iterations. Experimental results are provided to validate our findings.