PDF version of this document
Having argued for the validity of the method, we proceeded to a comparing of different methods of NRR. Of particular interest were group-wise and pair-wise method. Much as we had hope to show, it turned out that group-wise registration is superior to pair-wise. This assumption even holds over a wide range of parameter values that can be selected. This led to the much anticipated conclusion that our method can evaluate registration algorithms and little or no tweaking should be involved in advance so the method may be generic.
Current work appears to indicate that results of our evaluation agree with results that are based on ground truth. The advantage of our approach, on the other hand, is that is is data-driven and does not require additional information to be provided.