[Insight-users] Skeleton Filter?

[Andreas Krebs]

Hi Zihua Su

I already tried the filter proposed in the document
Digital Topology from Julien Lamy. It works fine, but the
terminality criterion needed to be improved for my application. I did it
and it works, but I programmed it in a very ugly way. So all in all
I have an ugly implementation of

Chris Pudney: "Distance-Ordered Homotopic Thinning:
A Skeletonization Algorithm for 3D Digital Images"
in: Computer Vision and Image Understanding, Vol 72, No. 3, 	
December, pp. 404--413, 1998

which works fine for images of 100 x 100 x 100. It becomes very slow for
bigger images.

If you already calculate the ChamferDistanceMap, then it is  worth to
ask, if you really need the skeleton or if the skeleton like ridges
which can be calculated (and implemented) much faster are sufficient.

For my application it was sufficient to look for the centers of maximal
disks (spheres). These can be identified from the chamfer distance
transform d(.) . The centers are the points p that satisfy

for all q \in N_26(p): d(p) \neq d(q) - Dist(p,q)

where Dist(p,q) is the distance between p and q (cf. Pudney, p. 408, eq.
(3) ). I use this approach also to find ridges in an grayscale image
based on fuzzy distance transform fdt(.). The output looks almost like a
skeleton.

If somebody would like to put these routines into the itk template
filter design, I would be glad. Perhaps my code can be helpful to
fulfill this task.

Furthermore, there is given a more complicated method to identify the
ridge using the Laplacian somewhere in the InsightApplicatiosn - I can
not find the link now. Anyway, I prefer my fuzzy distance transformation
ridges, after I played around with the Laplacian method.

Regards
	Andreas

Zihua Su schrieb:
> Hi all,
> 
> I have implemented the fastchamferdistancemap sucessfully. but i am
> wondering if there is a filter to take out the skeleton out in ITK!
> Any clue will be welcome and appreciated.
> 
> Cheers
> 

-- 
+--------------------------------------------------------------------------+
Andreas Krebs
Email: krebs.andreas at gmx.de
Digital signature certified by TC TrustCenter Class 1 CA
https://www.trustcenter.de/certservices/cacerts/TC_RootServer_DER_Class1.der
+--------------------------------------------------------------------------+