Fourier based registration of differentially scaled images

This paper presents a novel Fourier based approach for registration of images that have been translated, rotated, scaled and stretched. Fourier based techniques are computationally efficient and provide noise resilience but beyond similarity transformed images they do not extend readily to more general affine deformations. The proposed method relies on applying a Hough circle transform to a generated log-polar correlation surface in conjunction with various techniques for noise rejection. Applicable theory, methodology and preliminary results are presented.