PDF version of this entire document

Exploiting the Connection

There are two ways in which the connection between NRR and models can be exploited. NRR can be assisted by measures of model complexity and models can be built from NRR.

The main contribution of this thesis is the introduction of a generic method for assessing the quality of non-rigid registration []. The method does not require ground truth, but rather depends solely on the registered images. Consider the case where NRR is applied to a set of images, providing a dense correspondence between these images. Given this correspondence, it is possible to build a generative statistical model of appearance variation for the set. The quality of the resultant model will depend on the quality of the correspondence. The key idea that underpins the approach is that, if the correspondence is poor, the resulting appearance model will be poor. When the correspondences are correct, the model will be better. We define measures of model specificity and generalisation, which can be used to assess the quality of the model and, hence, the quality of the correspondence from which it is derived. This approach transforms the problem of assessing non-rigid registration to one of evaluating models generated from the results of registration. It does not depend on the specifics of the registration algorithm or the form of the model.