The most valuable contribution of this thesis is the introduction of a framework wherein NRR can be assessed without the use of ground truth. The framework is valuable not only because NRR applications are routinely used but because it offers a solution to a problem where cumbersome annotation is otherwise needed [].
The new assessment method exploits the relationship between NRR and model building, transforming the problem of assessing the NRR of a set of images into that of assessing the quality of a model built from the registered images.