Channel Modeling and Performance Analysis in Watermarking

Digital watermarking is equivalent to a communication problem, where the embedded message goes through an effective channel consisting of the image interference, noise, and attacks. Understanding and quantifying the performance of watermarking strategies, i.e., bit error rates (BER), under various scenarios is key to the further development of this field. Although a few watermark-specific and attack-specific results do exist, a general framework for performance evaluation of watermarking has not been available to date. Motivated by results in communication theory, we propose a modeling and performance evaluation strategy that uses the concept of intersymbol interference (ISI) channels and applies to a variety of embedding methods as well as various attacks. We first show the applicability of this general approach to several watermarks and attacks. We derive pair wise error probability equations that are used to calculate BER bounds. These bounds are verified by simulations. Finally, we study the use of MMSE detectors in watermarking, and give some results