Iterative Transform Parameter Refinement for Image Registration

Image registration is the process of determining the geometrical transformation that aligns two overlapping images. In this paper, a new automatic image registration system is presented for the estimation of 2D translation, rotation around the optical axis, and uniform scale differences between two overlapping images. Several aspects of the system, including fast corner extraction, rotationally invariant block matching, outlier suppression, similarity transformation estimation and parameter refinement using 2D clustering are discussed in detail. The performance of the system is demonstrated experimentally using various aerial images with different complexity.