Once we have registered the images, i.e., found matches and a rigid transformation, the question is: How confident can we be with this result? There are two main types of errors in feature-based registration algorithms. Firstly, the matches could be completely wrong and we simply recognized by chance n features in approximately the same configuration. This is called a gross error in statistics and a false positive in recognition. But even if we got the matches right, the features we are using to compute the registration are corrupted by noise and induce a small error, or uncertainty, on the transformation parameters. In this section, we analyze in turn these two types of error.
Was this article helpful?