Abstract
Feature matching between pairs of images is a main bottle- neck of structure-from-motion computation from large, unordered im- age sets. We propose an efficient way to establish point correspondences between al l pairs of images in a dataset, without having to test each individual pair. The principal message of this paper is that, given a suf- ficiently large visual vocabulary, feature matching can be cast as image indexing, sub ject to the additional constraints that index words must be rare in the database and unique in each image. We demonstrate that the proposed matching method, in conjunction with a standard inverted file, is 2-3 orders of magnitude faster than conventional pairwise match- ing. The proposed vocabulary-based matching has been integrated into a standard SfM pipeline, and delivers results similar to those of the con- ventional method in much less time.