[vtkusers] PCA deformation: How to recompose cells for maintaining shape "aspect"?

[Jonathan Bailleul]

Dear all,

I'm computing shape instances after a PCA analysis (using the VTK
filter) along a variation mode from -3 to +3 sigma. What I can see from
the pictures provided is that visualisation artefacts appear due to the
triangles connectivity.

In that (http://www.greyc.ismra.fr/~bailleul/These/m0.5sigma.jpg)
picture, my shape is not deformed that much, and the triangles
connecting landmarks (vertices here) 43, 45, 6, 26 form a good
approximation of what we should expect, ie a surface (nurbs?) that
should smoothly interplate these points.

But in that more deformed picture
(http://www.greyc.ismra.fr/~bailleul/These/p3sigma.jpg), we can see that
the movement of our points is showed as an obvious discontinuity (43 45
6) which prevents from having a good idea of what the shape should look
like. 

This is not an error from the model, since point ordering is purely
arbitrary fixed at the beginning: only points (landmarks) positions
matters. What I would like to make is to reorder cells so as to have (43
45 26) and (43 6 26) form 2 triangles, in that precise case.



I've taken time to look for filters into the documentation, but I've not
found anything of help. I would like to benefit from your experience,
since my problem might be solved from a way I didn't consider or by
realizing I've misunderstood something. Comments from people familiar
with PCA filter might be welcome in that context!

In advance, thank you for your help.


-- 
-----------------------------------
Jonathan BAILLEUL, Doctorant
GREYC Image - Université de Caen 
http://www.greyc.ismra.fr/~bailleul