A PCA-Based Binning Approach for Matching to Large SIFT Database

A method for efficiently finding SIFT correspondences in large keypoint archives by separating a database into bins – using the principal components of the SIFT descriptor vector as the binning criteria – is proposed. This technique builds upon our previous efforts to improve SIFT matching speed in image pairs, and will find correspondences approximately three times faster than FLANN – the approximate nearest-neighbor search library that implements the existing state of the art – for the same recall-precision performance.