Motion estimation based registration of geometrically distorted images for watermark recovery

Image registration is essential in spread spectrum based watermark detection of a distorted image. Traditional registration approaches include the use of templates and feature matching between the distorted image and a reference image. However, neither of these techniques adequately addresses localized transformation. We propose a new registration scheme based on motion estimation between the distorted image and a reference copy. Complex wavelets provide a hierarchical framework for our motion estimation algorithm and radial basis functions provide the means to correct erroneous motion vectors. Experimental results show that our approach can estimate the distortion quite accurately and allow correct watermark detection.