Registration of high resolution satellite images base on scale-orientation restricted KAZE

Image registration is an important preprocessing procedure for remote sensing image applications. In this paper a new multi-scale feature detector in nonlinear diffusion scale space KAZE is introduced to the registration of high resolution satellite images. We also presented an experimental evaluation for Scale Invariant Feature Transform (SIFT), Speeded Up Robust Features (SURF) and KAZE. KAZE shows the best performance with higher correct match rate and much more numbers of matched keypoint pairs. As a multi-scale approach, KAZE has a major shortcoming that a single local structure may produce redundant keypoints in a range of scale levels. In order to achieve robust feature matching in high resolution satellite images, we proposed a revised scale-orientation joint restricted matching algorithm for KAZE. Experimental results indicated that the proposed algorithm outperforms other matching methods in the correct match rate as well as aligning accuracy.