MSM: a scaling-based feature matching algorithm for images with large-scale differences

• Published in: International journal of digital earth Vol. 18; no. 1
• Main Authors: Ge, Qifeng, Du, Xiaoping, Xu, Chen, Xu, Jianhao, Yan, Zhenzhen, Fan, Xiangtao
• Format: Journal Article
• Language: English
• Published: Abingdon Taylor & Francis 31.12.2025; Taylor & Francis Ltd; Taylor & Francis Group
• Subjects: Algorithms; Datasets; Feature matching; Image quality; Image reconstruction; local feature; Matching; multi-scale dataset; multi-scale feature matching; Remote sensing; Satellite imagery; Simultaneous localization and mapping
• ISSN: 1753-8947; 1753-8955; 1753-8955
• DOI: 10.1080/17538947.2025.2543562

Feature matching represents a fundamental and critical problem in various tasks, including 3D reconstruction, simultaneous localization and mapping (SLAM) and preprocessing for remote sensing images. However, existing methods frequently fail to produce high-quality results on images with large-scale differences. In light of limitations, this study proposed a feature matching algorithm based on scaling, MSM (Multi-Scale Matching), which enhances feature matching performance in images with large-scale variations. This algorithm extract feature points across multiple scales, identifying them as scale-invariant keypoints. In addition, it combines descriptors from different scales to construct composite descriptors, to improve the robustness and accuracy of the matching process. For a more comprehensive evaluation of the MSM, we have meticulously devised a scale difference index (SDI) and constructed a multi-scale dataset (MSD) by SDI. The results of feature matching experiments demonstrated that the MSM algorithm outperforms current state-of-the-art methods regarding generality and effectiveness in several benchmark tests. This study presents a novel approach to feature matching in the context of large-scale differences, which can fulfill the requirements of feature matching for large-scale differences in UAV images, satellite images, etc. The MSD is accessible at https://github.com/KevenGe/MSD-Datasets .