Comparison of viewpoint-invariant template matchings

Abstract

SIFT, Forapro and Ciratefi are three rotation and scale invariant template matching (or keypoint matching) algorithms. They can become viewpoint-invariant using the view simulation technique, compensating the perspective deformation by simulated affine transformations, yielding the algorithms named ASIFT (Affine-invariant SIFT), Aforapro and Aciratefi. We compare the three algorithms, evaluating their performances in different situations. We focus our attention especially to situations with illumination changes and in the presence of repetitive similar patterns. The results show that all the three algorithms have strengths and weaknesses and the user should choose the best suited algorithm according to the intended application.