Watermark Embedding by Geometric Warping after Novel Image Moment-Based Normalization

Previous works have shown that the geometric structure of images and videos is a suitable embedding domain for watermark information. Using geometric warping data can be embedded very robust against even strong lossy compression. But, since the promising previous approach is based on a fixed pixel raster, the watermark extraction is fragile when rotation, scaling, translation, or shearing with unknown parameters is applied. To improve the previous work the image or video frame geometry has to be automatically resynchronized before watermark extraction. For that, in this paper, the geometric warping-based watermarking is extended by a novel image moment-based geometric normalization. The improved system is tested extensively and it is shown that the watermark data is robust against strong lossy image and video compression, noise, changes of luminance and contrast as well as scaling, rotation, translation, and shearing.