A Novel Approach to Image Roto-translation Estimation

An original approach to roto-translation estimation for images related by rotation and translation is investigated. The algorithm is operated in the frequency domain, where phase correlation is used for translation estimation and magnitude correlation is employed to rotation estimation. In the rotation estimation part, spectra of refereeing images are represented by Radon (projection) transform and their transformations are related by rotation angle. Based on this fact, an auxiliary function is defined and the problem is reduced to searching for the minimum of the function. Since the searching procedure is executed in projection domain, only some additions and few multiplications are involved in complexity, the computation burden is low comparing to classical phase correlation scheme. Experimental results demonstrate its applicability to roto-translation estimation.