Robust Multimodal Remote Sensing Image Matching Based on Enhanced Oriented Self-Similarity Descriptor

Abstract

The presence of complex geometric and radiometric differences in multimodal remote sensing images poses a significant challenge for existing image matching methods. In this letter, we propose a robust multimodal remote sensing image matching method based on an enhanced oriented self-similarity descriptor. Based on our previous oriented self-similarity descriptor, the assigned main orientations are refined by introducing a Gaussian smoothing function to improve robustness, and the redundant information is eliminated by downsampling the descriptor neighborhood to improve efficiency. Additionally, we design a position-weighted matching method to optimize the matching performance. Experimental results on three types of multimodal remote sensing image datasets comprising 75 image pairs demonstrate that the proposed method surpasses existing methods and exhibits superior robustness in achieving successful matching, with a success rate of 100 %. Our executable program and datasets are available for public download in https://github.com/Xxin08/EOSS.