Subpixel image registration algorithm based on pyramid phase correlation and upsampling

A fast subpixel image registration method is proposed in this paper. The implementation of this method is divided into two steps: coarse registration and fine registration. In the coarse registration stage, we propose a strategy to combine image pyramid with phase correlation; in the fine registration stage, we propose a strategy to perform local upsampling in the frequency domain through matrix multiplication. We compared our algorithm with traditional-feature-based and direct methods, as well as unsupervised learning algorithms. Our empirical results show that compared with traditional methods, our method achieves faster speed, while maintaining equivalent or better accuracy and robustness. In addition, compared with unsupervised learning algorithms, our method can be applied to real-time systems with higher speed requirements, better performance for cases with less overlapping regions, and better robustness to noise.