Image Registration–Implementation and Performance Analysis using Log-polar Transformation

Abstract

This paper presents an image registration algorithm for affine motion recovery. The images which are to be registered may be misaligned due to rotation, scale and translation the algorithm presented in the paper estimates affine transformation parameters to register this images. The parameters are computed using a variation of the Levenberg-Marquadt nonlinear least squares optimization method in a coarse-to-fine hierarchical framework. To precisely registers images with subpixel accuracy this method gives a robust solution. To accommodate arbitrary rotation angles and a wide range of scale changes a log-polar registration module is introduced. By using this method we achieve a good initial estimate for the optimization-based affine registration stage. This algorithm is implemented on pairs of digital images which are subjected to large affine motion.