Registration for Feature Point Sets Based on Affine Transformation

This work investigates the image registration from feature point sets. Image registration is a fundamental object recognition method in computer vision and it aims to find best matches between two or more point sets when there are geometric distortions, point measurement errors and contamination present. Up to now closed form solution has been developed only when the geometric distortion is similarity transformation. This paper concentrates on image registration from feature point sets when the geometric distortion between the two images is affine transformation and gives closed form solution for the transformation parameters that minimize the root-mean-squared residual error of the image points by the linear least-squares techniques and the pseudo-inverse of matrix respectively. In order to give the simple closed form solution, the image points are represented by homogeneous coordinates and the theories of matrix are used. The algorithms are evaluated on both synthetic and real world images and the experiment results show that the methods given in this paper are accurate, stable and are only affected slightly by noise.