Robust data hiding with image invariants

We investigate data hiding with Zernike moments (ZMs) of an image, in an attempt to achieve geometric robustness by drawing on the fact that the magnitudes of ZMs are invariant under image rotation and flipping. However, with the conventional Cartesian method for ZM computation, the invariance property is far from ideal. We propose a polar coordinate based approach to compute ZMs more accurately, and thus to improve the invariance property. The data is embedded into an image by modifying slightly the magnitudes of its ZMs with dither modulation. Experimental results show that the watermark based on the proposed algorithm has better robustness to geometric attacks including image rotation, scaling, flipping, as well as a number of other common manipulations such as lossy compression, low pass filtering