[Insight-users] label object manipulation

[Padfield, Dirk R (GE, Research)]

Hello Gaëtan,

Thanks a lot for your message.  As you mentioned, the itkLabelGeometryImageFilter does not yet take pixel spacing into account, and I need to find time to make this extension.  This extension should, however, be fairly straightforward.  For the problem of including the voxel thickness in the moment calculation, I believe the derivation of Equation 20 in the itkLabelGeometryImageFilter documentation (http://insight-journal.com/download/viewpdf/301/4/download) should be useful since it is derived by analytically calculating the moment of a pixel.  It should be easy to generalize this derivation to take into account pixel spacing.

I hope this helps,
Dirk.

 

-----Original Message-----
From: Gaëtan Lehmann [mailto:gaetan.lehmann at jouy.inra.fr] 
Sent: Tuesday, June 30, 2009 5:00 AM
To: lynx.abraxas at freenet.de; Padfield, Dirk R (GE, Research)
Cc: insight-users at itk.org
Subject: Re: [Insight-users] label object manipulation


Le 29 juin 09 à 23:31, lynx.abraxas at freenet.de a écrit :

> Dear Gaëtan Lehmann,
>
>
> Thanks  a  lot  for You quick answer. With Your hints it was easy to 
> extend my little program to suit my needs further.
> I'm wondering though if the article is the best documentation to work 
> with.  I
> couldn't   find   the   Perimeter   attribute   only   
> EquivalentPerimeter  and
> EquivalentEllipsoidPerimeter. The connectivity is also only mentioned  
> in  the examples.  Is there another documentation where these are 
> explained and that I should read before asking here?
>

No, there is no other document, but you can look at the source! All the attributes are listed in

   http://voxel.jouy.inra.fr/darcs/contrib-itk/binaryAttributeMorphology/itkShapeLabelObject.h
   http://voxel.jouy.inra.fr/darcs/contrib-itk/binaryAttributeMorphology/itkStatisticsLabelObject.h

And you're right, I forgot to put the perimeter in the attribute list.  
I'll try to fix that in a next revision.

> In the footnote it says there'll be more about perimeter estimation in 
> another article. Is that published already?
>

No, not yet.

> I  noticed  that GetEquivalentEllipsoidSize yields nan/inf if the 
> bounding box is only 1 pixel wide. When I used the image moments of 
> Avizo to calculate  the axes lengths a bounding box width of 1 pixel 
> wasn't a problem. I'm not sure if
> this result from Avizo is good but with my current understanding a   
> full  axis
> length of 1 (or 0.5 for a semi axis) would be sensible, wouldn't it?

I have this problem also, and it's quite annoying. It's caused by an approximation in the moment computations: the voxel thickness is not used.
The formulae I used are very common, and I didn't find an easy way to include it. In http://insight-journal.com/browse/publication/301 , this information is used, but unfortunately, the spacing is not.
Also, I don't think that putting 1 (or 1/2) axis if the size is 0 on this axis would be nice, because it would make the result quite different of the case where there the size is very thin but not zero.  
Something like

    0000000000000000000000000000

and

    0000000000000000000000000000
           0


> If  it's not possible to calculate that, would it be possible to get 
> an output letting the filter assume only 2D for these objects?


I would prefer getting a non-approximated formulae for the moment computation which take care of the spacing. I'm quite sure it's not that complicated, but, well, that's not my field of competences.

If someone has some documentation, or has the needed knowledge on that point, I would really appreciate some help!

Gaëtan




--
Gaëtan Lehmann
Biologie du Développement et de la Reproduction INRA de Jouy-en-Josas (France)
tel: +33 1 34 65 29 66    fax: 01 34 65 29 09
http://voxel.jouy.inra.fr  http://www.itk.org http://www.mandriva.org  http://www.bepo.fr