Abstract

In this paper we study interest point descriptors for im-
age matching and 3D reconstruction. We examine the build-
ing blocks of descriptor algorithms and evaluate numerous
combinations of components. Various published descriptors
such as SIFT, GLOH, and Spin Images can be cast into our
framework. For each candidate algorithm we learn good
choices for parameters using a training set consisting of
patches from a multi-image 3D reconstruction where accu-
rate ground-truth matches are known. The best descriptors
were those with log polar histogramming regions and fea-
ture vectors constructed from rectified outputs of steerable
quadrature filters. At a 95% detection rate these gave one
third of the incorrect matches produced by SIFT.

‚Äč