Shape Correspondence Based on Local Curvature
We are working on a correspondence method based on local curvature. This is a population based method (a la MDL or DetCov) that optimizes correspondence for a population, rather than pair-based. Currently we use the Koenderink measure C and S as our correspondence metric. However, the code is structured in such a way to enable very easy modification of the metric.
Current Status & Plans for Project Week
There still is plenty of room for improvements, but the code is working right now. We have tried it in a very small dataset for debugging, and the optimization seems to properly converge.
We also tried on the caudate dataset, and it seemed to work properly, although we dont yet have a visualization method in place. This is one of the things we want to accomplish at Project Week: coming up with a nice visualization of the correspondence results with the KWMeshVisu tool that we developed at the previous programming week.
We are also in contact with Tobias Heimann who is the original author of a large part of the ITK classes we are using, and we plan to make a combined submission with him to MICCAI open source workshop. The paper is also one of the major things that we want to work on during the week.
- Ipek Oguz (UNC)
- Martin Styner (UNC)