Robust surface matching for registration

Abstract

The ability to accurately register images of patients taken at different times is very important in many medical applications. In most cases, the main reason for registration is to measure changes, and therefore any automatic registration algorithm employed for this task must be able to cope when there are significant differences in the images. In this paper a new algorithm is presented which enables accurate surface registration by using a robust matching algorithm to calculate correspondences for all the points on the surface along with a measure of reliability. The basic idea is to move the point of interest to see if it always matches to the same point. This generates a scatter of "tentative corresponding points", and a corresponding point and reliability are calculated from the geometric distribution of these points. This new algorithm is compared with two others. Experiments show that our algorithm is more accurate, consistent and converges faster than the others.