Affine transformation invariant image watermarking using moment normalization and radial symmetry transform

A novel technique for image watermarking aiming at robustness against geometrical attacks is proposed in the current paper. The first stage relies on computing a normalized version of the original image using image moments. A radial symmetry transform is then applied to the normalized image, followed by a non-maxima suppression step. The feature points extracted this way act as centers of the area where the watermark is to be embedded. To minimize image distortion, we embed the inversely normalized watermark (initially formed in the DCT domain) in the original image. The detection process consists of normalizing the input image, applying the radial symmetry transform and extracting the feature points, after which a correlation detector is employed to detect the possibly inserted watermark in the normalized image. Experimental results prove the robustness of the method against affine geometrical transform attacks, as well as other image processing operations.