A Multi-Modal Image Matching Approach Based on Nonlinear Anisotropic Filtering and GMS

Nonlinear Radiometric Distortions (NRD) between images from different sensors will lead to image matching difficulties. The KAZE algorithm is a robust method to NRD effect between multi-modal images, but the matching quality is sensitive to the parameter settings. To solve this problem, a novel feature matching algorithm that is robust to parameter settings is proposed in this work. A novel conduction function based on the KAZE algorithm is utilized, enabling better preservation of edge information during the filtering process. Additionally, the integration of an enhanced GLOH descriptor improves the computational efficiency of the proposed method. In the matching stage, the RANSAC algorithm is applied alongside GMS motion smoothing constraints to effectively eliminate mismatched points. Different experiments with three groups of satellite images are conducted to verified the effective of the proposed method.