PDF version of this document

Overview

Non-rigid registration (NRR) is used increasingly routinely in medical image analysis. There are, however, many different approaches to NRR, each leading to different results. We wish to find an objective method for evaluating the quality of NRR, so that different approaches can be compared and the quality of the NRR step in real applications can be be assessed. The main contributions of the work are:

• objective evaluation of NRR;
• no ground-truth required;
• validation using perturbed NRR;
• comparison with ground-truth based approach;
• practical comparison of NRR algorithms.