As demonstrated by Jin et al., the traditional machine learning algorithm using the scale-invariant feature transform (SIFT)33 and random sample consensus (RANSAC)34 may outperform the state-of-the-art DL methods for image matching35.