Image Registration of CAS500-1 and KOMPSAT-3/3A Satellite Images Using Deep Learning-Based Feature Matching and Super-Resolution Techniques

Satellite imagery is essential for continuously monitoring Earth phenomena, detecting disasters and hazards, and effectively identifying large and small-scale changes across wide areas. Over the past few decades, advancements in satellite technology have significantly increased the use of satellite imagery. In particular, in change detection studies or disaster monitoring research utilizing multi-temporal and multi-satellite imagery, the fusion of images from two or more time periods for the same region is indispensable. However, due to the inherent characteristics of satellite imagery being captured from a distance, geometric distortions are likely to occur, potentially resulting in misalignment between the images and the actual ground surface. The accuracy of high-resolution satellite imagery is determined by the precision of geometric corrections, which becomes an even more critical factor when using multi-satellite and multi-temporal imagery. Consequently, image registration is an essential process in studies that utilize the fusion of high-resolution satellite imagery. In this study, we propose a highly accurate image registration method using high-resolution satellite imagery from CAS500-1, KOMPSAT-3A, and KOMPSAT-3. To overcome the limitations of feature point detection, a ResShift-based super-resolution technique was applied to generate a dataset with higher resolution than the original data, maximizing the performance of the feature matching models. For deep learning-based feature point detection and matching models, SuperPoint, SuperGlue, LightGlue, and RoMa were utilized. Notably, the RoMa model demonstrated exceptional performance by recording over 2,300 correct matches on the super-resolved dataset. The results of this study are expected to contribute to effective image registration in various fields that utilize multi-temporal and multi-satellite imagery.

This work is supported by the Korea Agency for Infrastructure Technology Advancement(KAIA) grant funded by the Ministry of Land, Infrastructure and Transport (Grant RS-2022-00155763).