Hierarchical hybrid multi-scale feature match

摘要

Finding reliable corresponding points between two images of a scene is a fundamental problem in computer vision. In this paper, a hybrid scheme is proposed, which combines invariant spatial feature and frequency domain based methods in a hierarchical multi-scale way. The Fourier-Mellin Transform is applied to obtain the transformation parameters at the coarse level between the two images; then, the parameters can serve as the initial guess, to guide the following feature matching step at the original scale, where the correspondences are restricted in a search window determined by the warp between the reference image and the current image; Finally, the transformation parameters are refined by a RANSAC procedure. This in return provides a more accurate result for feature correspondence. Experiments show that our approach provides satisfactory feature matching performance. This method also makes precise geometric rectification to remote sensing imagery.