[Insight-users] is the ITK consistent regarding signed distance functions ?

[Oliver Gloger]

Hello,

I would like to use the GeodesicActiveContourLevelSetImageFilter and refine its result with the CannySegmentationLevelSetImageFilter. (2D-images first!!!)

Hence I combined the Example-Code for those 2 filters:

I passed the output of the GeodesicActiveContourLevelSetImageFilter as initialModel (initial signed distance function) to the CannySegmentationLevelSetImageFilter.

However, I wondered If I look at the ouputs (I visualized them via png-images) of the FastMarchingImageFilter and the GeodesicActiveContourLevelSetImageFilter: they have lower (I assume:negative) values inside and higher values (I assume positive) outside. Contrarily to the convention that is used in the ITKSoftwareGuide (Fig.9.13)!!!

Hence, the ouputs (after BinaryThresholding given in the example-code) of the GeodesicActiveContours are quite well (inside =="white", outside=="dark").  But the subsequent binary-outputs of the CannySegmentationLevelSetImageFilter are different (inside =="dark", outside=="white")! And the results are worse than the results of the GeodesicActiveContourLevelSetImageFilter.

So, my question is: 

What is the convention in the ITK for signed distance functions? Inside the zero level set >0 like in Fig. 9.13 or the opposite? Is the ITK not consistent here??????

Or did I make another mistake? 

Best regards

Oliver

______________________________________________________
GRATIS für alle WEB.DE-Nutzer: Die maxdome Movie-FLAT!
Jetzt freischalten unter http://movieflat.web.de