[Insight-users] ICP

[Luis Ibanez]

Hi Anne-Laure,

For the large number of points that you have, it is better to avoid
using ICP directly, due to the large amount of time that it takes
to compute the distances between the two sets of points.

This, plus the fact that ICP itself is patented, are enough reasons
for exploring variations of the algorithm.

In your case, since the Fixed points come from a segmentation of the
rib case, and the moving points come from a theoretical model, it is
worth to explore if for one set or the other you could use a distance
map.

For example,
Is is possible to generate a distance map to the rib cage model ?
instead of getting 8435 points out of it ?

If so, then you could register the moving point cloud to the distance
map of the rib cage model.

You could also try to other way: generate a distance map to the
segmentation of the rib cage, and register the points of the theoretical
model to the points in the segmentation.

Combining PointSet + DistanceMap is probably the most efficient
way of computing point set registration because you take advantage
of the efficiency in finding the "closest distance" when you query
a DistanceMap, and you take advantage of the small number of samples
that (in general) you have in a PointSet.


Note that the PointSet to DistanceMap registration can be performed
with existing ITK classes.



   Please let us know if you have further questions,



       Thanks


         Luis


------------------
annelaure wrote:
> Thank you so much for your help in all the problem I met with ITK.
> My register problem :
> I have two points clouds with different number of points (fixed points :
> 89680, moving points : 8435). Fixed points com from the segmentation of the
> rib cage (points cloud non regular and with holes so not usable just as it
> is) and moving points come from a theorical model of a rib cage. The aim of
> my work is to register my theorical model on the real rib cage in order to
> have a personalized model of the thorax.
> I don't know which method of register is the better for me.
> Thank you so much,
> regards,
> Anne-Laure