A Multi-temporal Image Registration Method Based on Edge Matching and Maximum Likelihood Estimation Sample Consensus

In the paper, we propose a newly registration method for multi-temporal image registration. Multi-temporal image registration has two difficulties: one is how to design matching strategy to gain enough initial correspondence points. Because the wrong matching correspondence points are unavoidable, we are unable to know how many outliers, so the second difficult of registration is how to calculate true registration parameter from the initial point set correctly. In this paper, we present edge matching to resolve the first difficulty, and creatively introduce maximum likelihood estimation sample consensus to resolve the robustness of registration parameter calculation. The experiment shows, the feature matching we utilized has better performance than traditional normalization correlation coefficient. And the Maximum Likelihood Estimation Sample Conesus is able to solve the true registration parameter robustly. And it can relieve us from defining threshold. In experiment, we select a pair of IKONOS imagery. The feature matching combined with the Maximum Likelihood Estimation Sample Consensus has robust and satisfying registration result.