| Image registration means alignment of two or more images. Among a large number of image registration techniques, maximization of mutual information (MMI) has become the most popular method for parametric image registration problems. As implied by its name, (global) maximization is the major component of this technique. The purpose of this study is to design a robust yet efficient global optimizer based on the DIRECT algorithm and a heuristic global optimum test algorithm for parametric image registration problems using mutual information as the similarity measure to be maximized. DIRECT algorithm is a deterministic global optimization algorithm. It works by systematic exploration of sub-regions in the region of interest. Like most of the existing global optimization algorithms, DIRECT algorithm needs an effective stop criterion to terminate. In this thesis, we employ a heuristic global optimum test algorithm as the stop criterion and show how to integrate them effectively. The efficacy of the developed optimization scheme is illustrated by registering several pairs of remote sensing images in 2D and 3D cases. |
