The RST invariant digital image watermarking using Radon transforms and complex moments

Many digital image watermarking techniques are sensitive to geometric distortions. These distortions can destroy the synchronization between the watermark and the watermarked image to make it difficult or even impossible to detect and extract the embedded watermark. In this paper, we first present an approach of rotation, scaling, and translation (RST) invariant analysis for images. This approach achieves a set of completely invariant descriptors from the complex moments of the original image's Radon projection. Then, we propose a watermarking scheme which can resist global geometric transforms. Watermark embedding is performed in discrete cosine transform (DCT) domain of the original image. The watermarking detection is as simple as computing a set of the proposed invariant features to illuminate the robustness problem under geometric image manipulations. In the extraction scheme, we first estimate the geometric distortion factors by using moments of watermarked image. With these estimated factors, the geometrically transformed image can be converted back into its original position, size and shape, to recover the lost synchronization for watermarking extraction. We systematically evaluate the proposed method using several experiments. The results demonstrate that the embedded watermark has a good robustness to not only the translation, rotation, scaling and aspect ratio changes, but also some common image processing operations such as additive noise and etc. Experiments also indicate that the proposed watermarking scheme has advantages over the other similar ones in detection and extraction aspects, and exhibits the capability in the image authentication and the copyright protection.