A novel optical-SAR image registration method based on KAZE features
Yu Zhang, Qiming Zhang, Z D Zhang, Jinping Sun · IET conference proceedings. · 2026
The registration technology of optical and synthetic aperture radar (SAR) images constitutes the basis of image fusion. It is important for improving the precision of target detection and location. Due to the difference in imaging principles, there is severe nonlinear radiation distortion between optical and SAR images, limiting the matching performance of existing registration algorithms. In this paper, we propose a novel registration method based on KAZE features for optical and SAR images. According to the unique characteristics of optical and SAR images, the proposed method employs nonlinear diffusion filtering and multi-scale Sobel operators to establish nonlinear scale spaces for SAR images and Harris scale space for optical images, improving the accuracy of feature detection. Moreover, the detected feature points are described by using the gradient location and orientation histogram in multiple areas, improving the distinctiveness of feature description. The optical-SAR image registration experiment has verified the matching performance of the proposed method.