Efficient SIFT matching from keypoint descriptor properties

A modular approach to finding fast SIFT correspondences in single-image matching applications is proposed. Our algorithm exploits properties of the SIFT descriptor vector to find shortcuts to the most likely matches in a feature set. We are able to converge approximately 15 times faster than a linear search, and, respectively, four and five times faster than both PCA-SIFT and SURF (both of which use descriptor vectors that contain far fewer dimensions than SIFT), at near-equivalent recall and precision performance.