Robust Watermarking in DoG Scale Space Using a Multi-scale JND Model

In this paper, a novel watermarking method in the Difference of Gaussian (DoG) Scale Space is proposed. The idea is to decompose image into DoG scales and insert the watermark into these DoG sub-images using a multiscale JND (Just Noticeable Difference) model, providing an invisible and robust watermarking scheme. In order to survive de-synchronization attacks, we use the SIFT (Scale Invariant Feature Tranform) keypoints detection. Both keypoints detection and JND mask are performed in the DoG scale space, reducing then the complexity of the method. An intensive experimental evaluation is carried out to demonstrate that the proposed technique is transparent and robust to a wide variety of attacks from “signal processing” to de-synchronization type, especially severe attacks like Print-Scan and Camcorder.