Conclusions
Our results suggest that the performance of NNR algorithms can be compared objectively, using just the registered images they produce, and that the quality of registration can be assessed in routine applications of NRR, without the need for any additional information.