PDF version of this entire document

Contributions

The contribution of this work is two-fold:

• On the one hand, it is demonstrated that one is able to evaluate the quality of NRR without requiring ground-truth data. Thus, a comparison between various NRR algorithms can be made without cumbersome-to-obtain manual annotation.
  • Further to this, experimental validation is used to show that the method works.
  • Several registration algorithms are compared using the new method, demonstrating the use of this method in a practical setting.
• On the other hand, one is also able to build models automatically, using registration algorithms, and then evaluate the resultant models.

Overall, this provides a framework for automatic or semi-automatic analysis of arbitrarily-large amounts of data.