[Insight-users] PointSet Segmentation

[Karthik Krishnan]

On Thu, 2005-08-11 at 16:26 -0400, Kevin H. Hobbs wrote:
> On Thu, 2005-08-11 at 15:12 -0400, Karthik Krishnan wrote:
> > 
> > Kevin H. Hobbs wrote:
> > 
> > >Is there any way to segment PointSets or Blobs?  I looked through the
> > >software guide, and the Doxygen, and I didn't see anything.
> > >  
> > >
> > If your point set is sparse enough, you can use a clustering algorithm.
> > 
> > You can use Kmeans clustering from ITK.
> > Take a look at Examples/Statistics/KdTreeBasedKMeansClustering.cxx
> > 
> > Your sample in the example will be the list of points in the point set. 
> > You can define the number of clusters you want.
> > Since Euclidian distance metric is used, you will end up binning your 
> > sparse image into n clusters.
> > 
> > HTH
> > karthik
> 
> From a 10 min read of the software guide, it sounds like k-Means
> clustering would work well for sphere-ish things.  Is this correct?  

True.

> The
> things I want to segment are mostly long, skinny, and curvy.  I suspect
> this would cause trouble. 

Normalized cuts or one of the graph cuts algorithms might help here. One
of both of them will be in the toolkit very soon. You may also want to
look at the MeanShfit methods in ITK. I am not familiar enough with them
to comment on them.

thanks