Evaluation of Similarity Measures for Non-rigid Registration

摘要

In this paper we propose a protocol for the evaluation of similarity measures for non-rigid registration. The protocol is feasible for the evaluation of non-rigid registration in which the deformation model is based on a set of regularly or irregularly distributed corresponding point pairs. The proposed protocol is able to deduce five properties of the similarity measure for each point pair of the deformation model, so that local or global estimation of the similarity measure properties can be derived. The feasibility of the proposed protocol was demonstrated on a popular deformation model based on B-splines, on six similarity measures, and on the "gold standard" CT and MR images of three spine vertebrae and three MR T1 and T2 images of the head.