Detection of rotation and parallel translation using Hough and Fourier transforms

Abstract

This paper proposes a method of, given a multilevel image called a reference image in which an arbitrary shape S lies and another multilevel image called an input image in which one shape rotated and parallel translated from S lies, computing the angle of rotation and the quantity of parallel translation. This method applies a modified version of the Hough transform proposed by O'Gorman and Clowes (1976) to the reference and input images, and computes the angle of rotation uniquely. Moreover, this method needs no extracting feature points, and is applicable to low quality images.