Rotation, Scaling, and Translation Resilient Image Watermarking with Side Information at Encoder

We propose a rotation, scaling and translation (RST) resilient blind image watermarking technique by using Fourier-Mellin transform and informed coding, where watermark detection does not require the existence of the original image. Although the Fourier-Mellin invariant domain is recognized to be nearly RST invariant, which supposedly would make it an ideal space for watermark embedding, however the inverse Log-Polar map (ILPM) could severely destroy the embedded watermark. Therefore a very strong watermark has to be employed to survive the self-destruction process from ILPM, but in turn this leads to a poorly watermarked image. To overcome this problem we introduce the concept of informed coding by embedding a slightly weaker watermark into the image so that both transparency and robustness can be achieved. Experimental results demonstrate that the resulting watermark is robust to a variety of image processing attacks.