A Robust Feature Point Matching Method for Dynamic Aerial Image Registration

Abstract

Feature matching is a critical and challenging process in feature-based image registration. In this paper, a robust feature point matching method, combined Scale-Invariant Feature Transform (SIFT) and Support Vector Machine (SVM), is proposed to match features for registering dynamic aerial images. In this method, every feature point is described by 128 dimensional SIFT descriptor as a training vector. Then feature matching model is built by SVM. Using this model, feature points are classified into two categories, one is matched feature set and the other is unmatched feature set. Three pairs of infrared (IR) and ultraviolet (UV) aerial images are utilized to evaluate the performance. The matching results have confirmed that the proposed method can match the feature points exactly even with a lot of outliers.