Remote Sensing Image Registration based on Multi-feature Selection Strategy
Qiang Chen, Lu Chen, Qing Zhao, Tuwen Huang, Hongrui Gong · 2024
With the rapid development of remote sensing technology, remote sensing image registration has been fully developed in various fields. However, for multi-temporal and multi-scene remote sensing images, due to the differences in imaging conditions, the existence of feature changes between images and other issues, there will be differences in image quality, spectral characteristics and spatial resolution. This brings great challenges to the work of remote sensing image registration. In order to better align remote sensing images, we propose an automatic registration method for multi-type remote sensing images based on multi-feature selection strategy. First, for multi-view and multi-temporal remote sensing images, this paper utilizes the nearest neighbor ratio-based SIFT algorithm for feature point extraction, and second, for multi-sensor types, this paper proposes a feature descriptor optimized based on the Histogram of Oriented Gradient HOG (HOG) and incorporating a new thresholding strategy, and we refer to this new method as simply as Direction Gradient Feature (DGF). Next, the matching set of feature points is determined by a multi-feature selection strategy. Finally, a hybrid Gaussian model is used to calculate the correspondence and complete the registration between multi-temporal and multi-modal images. The experimental results show that this algorithm combines the advantages of the two algorithms to improve the comprehensiveness of the algorithm and significantly improve the image registration accuracy.