Robust image watermarking against local geometric attacks using multiscale block matching method

This paper presents an image registration algorithm against the local geometric attack which is recently considered as the public enemy of most watermarking schemes. The watermark is embedded and retrieved using an amplitude modulation watermarking scheme. This work gives consideration to all the fidelity, robustness, and capacity. Moreover, the watermark bits act as implicit anchors to help correct local geometric distortions. In the proposed registration algorithm, an automatically multi-scaled neighborhood-matching approach is used to quickly and efficiently estimate local displacement, and a multi-resolution framework is superimposed to estimate both small and large local geometric distortions. Experimental results demonstrate the robustness and efficiency of the proposed algorithm.