[ITK] Questions about ITK-Snap gaussian smoothing

[Beyer, Benedikt Alexander]

Hi Dženan and Paul,

thank you for the Information about the mesh smoothing algorithm, that helped a lot. Unfortunately I don’t completely understand the source code, since I’m not a software developer.

In the source code you linked it says that the maximum error parameter is ignored, so I assume that the value in the program settings has no effect on the smoothing.
(https://sourceforge.net/p/itk-snap/src/ci/master/tree/Logic/Mesh/VTKMeshPipeline.cxx#l169)

As for the sigma I still don’t entirely understand why the polygon model is not smoothed below a sigma of 0.67.
In the preferences for ITK-SNAP at the 3D Rendering tab, there is a Gaussian image smoothing option. Activating this option will smooth the polygon model of a segmentation for sigma>=0.67.

Also in the source code and documentation there seems to be contradicting information, because in vtkImageGaussianSmooth.h<https://gitlab.kitware.com/vtk/vtk/blob/master/Imaging/General/vtkImageGaussianSmooth.h> sigma is in pixel units and in VTKMeshPipeline.cxx#l166<https://sourceforge.net/p/itk-snap/src/ci/master/tree/Logic/Mesh/VTKMeshPipeline.cxx#l166> it is in millimeters. Which information relates to the 3D Rendering option?

Regards,
Benedikt Beyer

Von: Dženan Zukić [mailto:dzenanz at gmail.com]
Gesendet: Dienstag, 30. Mai 2017 15:38
An: Beyer, Benedikt Alexander <benedikt.beyer at rwth-aachen.de>
Cc: community at itk.org; Paul Yushkevich <pyushkevich at gmail.com>
Betreff: Re: [ITK] Questions about ITK-Snap gaussian smoothing

Hi Benedikt,

if you are asking about which mesh smoothing algorithm is used, I think this source code file answers that:
https://sourceforge.net/p/itk-snap/src/ci/master/tree/Logic/Mesh/VTKMeshPipeline.cxx#l167

Its declaration<https://sourceforge.net/p/itk-snap/src/ci/master/tree/Logic/Mesh/VTKMeshPipeline.h#l122> says it is an vtkImageGaussianSmooth<http://www.vtk.org/doc/nightly/html/classvtkImageGaussianSmooth.html>. If documentation is not enough, you can find its source code<https://gitlab.kitware.com/vtk/vtk/blob/master/Imaging/General/vtkImageGaussianSmooth.cxx> in VTK.

As the sigma is passed into there as a parameter, I believe the user-modifiable preferences in the program control it.

Paul (CC'd) might have more explanations.

Regards,
Dženan Zukić, PhD, Senior R&D Engineer, Kitware (Carrboro, N.C.)

On Tue, May 30, 2017 at 4:42 AM, Beyer, Benedikt Alexander <benedikt.beyer at rwth-aachen.de<mailto:benedikt.beyer at rwth-aachen.de>> wrote:
Dear ITK community,

I am currently writing an academic thesis at the RWTH Aachen University and I work with STL-data that is exported with ITK-Snap. A segmented structure is exported as an STL-file and then analyzed with VG Studio MAX. The data is used to determine measurement uncertainties created by gaussian filtering. To determine this uncertainty, I compare smoothed data to unsmoothed data.  I have a few questions regarding the gaussian image smoothing setting.


  1.  What gaussian filter algorithm does this option use? Eg. Recursive, discrete, robust, etc.
  2.  Is this filter a linear or surface filter?
  3.  How big is the gaussian kernel? Eg. 3x3, 5x5, etc.
  4.  What does the “approximation error max” option do? It has no visual effect on the segmentation.
  5.  There seems to be a threshold of the standard deviation at around 0.67. It seems that below this number no smoothing is applied. Is this specific to the program, or is this dependend on the voxel size?
  6.  Are there any known influences on measurement uncertainty by gaussian smoothing?

Thank you in advance and with kind regards
Benedikt Beyer



_______________________________________________
Community mailing list
Community at itk.org<mailto:Community at itk.org>
http://public.kitware.com/mailman/listinfo/community

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20170613/2c2b9d5d/attachment.html>