A High Precision Feature Matching Method Based on Geometrical Outlier Removal for Remote Sensing Image Registration

The reliability of feature matching is required for accurate remote sensing image registration. Outliers reduce the accuracy and effectiveness of feature matching. This paper proposes a novel feature matching method that utilizes the global geometric relationship of feature points in two images to eliminate outliers and preserve inliers. A mathematical model is formulated based on the similarity of the geometric relationship of feature points in the reference image and the sensed image. We also find the optimization solution through analysis and simplification of the mathematical model. The corresponding feature matching algorithm based on outlier removal is proposed according to the optimization solution. The experimental results of several remote sensing images demonstrate that our method can preserve more inliers, remove more outliers and obtain a better registration performance with higher accuracy and robustness than the state-of-the-art methods, such as SIFT, SIFT-RANSAC, SIFT-GTM, SIFT-LPM.