[Insight-users] Re : Re : Re : active contour, snake, deformable model and level set segmentation ???

Mon Dec 15 05:08:01 EST 2008

Hi Luis, 
Thank you for your suggestions. I will give a try and hope and get some good results.
Thank you
Best regards 
Juliette

________________________________
De : Luis Ibanez <luis..ibanez at kitware.com>
À : Juliette Deniau <juliette.deniau at yahoo.fr>
Cc : insight-users at itk.org
Envoyé le : Mercredi, 10 Décembre 2008, 19h25mn 59s
Objet : Re: Re : Re : [Insight-users] active contour, snake, deformable model and level set segmentation ???

Hi Juliette,

Thanks for the additional information about the problem that you are
trying to solve.

LevelSet filters may not be the ideal tool for segmenting objects with
high curvatures. At least not the GeodesicActiveContour nor the Shape
Detection level set implementations.

You may want to give it a try at:

  a) ThresholdSegmentationLevelSet
  b) FastMarchingLevelSet

(a) Will depend strongly on the intensity levels of the input image,
    and therefore will behave close to the Region Growing filters,
    (ConfidenceConnected and ConnectedThreshold).

(b) Will depend strongly on the gradient magnitude of the input image.
    That is, if your input image have well-defined edges around the
    objects of interests (the weapons) then the segmentation will be
    well contained.

    Otherwise, the segmentation will tend to "leak" in regions where
    the edge has low contrast.

If you are doing this from CT images, you probably should try to take
advantage of the fact that the Hounsfiel Units are somehow normalized.
However, presumably, the objects of interest will be surrounded by objects of great diversity, and it may not be possible to count on
having a good contrast against the background.

I'll suggest you to give it a try to the two filters above, but be
aware that you still will have to fine tune a set of parameters...

    Regards,

        Luis

----------------------
Juliette Deniau wrote:
> Thank you Luis for your massage and patience!
> 
> Sorry for my question, I'm just a student not an expert! But according to my reading in active contour and level set papers, I have found most of papers that are combining somehow level set with active contour which means that are not the same!
> 
> Any way what I would like to segment as object are weapon objects (knifes  and guns from data base of still images(2D)).
> 
> These objects are usually small and thin in particular for the knife objects.. I have already tested the region growing (Confidence connected filter), it gives acceptable results but it depends highly in the parameters, in particular the factor parameter.
> 
> Now I try to use level set and active contour based segmentation since they are more robust than the region growing methods.
> 
> Hope I have well explained my problem and hope I can get help and guidance from you. 
> Many thanks.
> 
> Regards 
> Juliette
> 
> 
> 
> ------------------------------------------------------------------------
> *De :* Luis Ibanez <luis.ibanez at kitware.com>
> *À :* Juliette Deniau <juliette.deniau at yahoo.fr>
> *Cc :* insight-users at itk.org
> *Envoyé le :* Jeudi, 4 Décembre 2008, 19h46mn 27s
> *Objet :* Re: Re : [Insight-users] active contour, snake, deformable model and level set segmentation ???
> 
> 
> Hi Juliette,
> 
> 
> I'm not sure what you mean by:
> 
> 
>  "Level set (purely level set without active contour)
>  based segmentation?"
> 
> 
> The level set abstraction is implemented in such a way that at
> every pixel ITK solves an increment of a partial differential
> equation.
> 
> 
> The surface of the object happens to be defined at the loci
> where the level set image has zero values.
> 
> 
> 
> 
> In practice, the distinction between a LevelSet and an Active
> Contours (or Snakes) is mostly of importance for those researchers
> who need to complete their annual quota of published papers and
> therefore are forced to be inventing "new" things every time, even
> when they are reincarnations of older ideas.
> 
> 
> In essence, the difference between LevelSets and Active contours
> is one of software implementation.
> 
> 
> If you need to solve a real and practical problem, we could
> advice you on what ITK methods to use.
> 
> 
> If you are in need of a flashy name for a segmentation method
> then...  there is a lot out there to choose from.  :-)
> 
> 
>  Regards,
> 
> 
>      Luis
> 
> 
> ----------------------
> Juliette Deniau wrote:
>  > Hi Luis,
>  >
>  > A one more question to clear completely my understanding.
>  >
>  > You said that “The Level Set implementations of deformable models have all
>  > the "LevelSet" string as part of their name” . Does this means that there is no
>  >
>  > level set (purely level set without active contour) based segmentation?
>  > Thank you for your time
>  >
>  > Juliette
>  >  >
>  > ------------------------------------------------------------------------
>  > *De :* Luis Ibanez <luis.ibanez at kitware.com <mailto:luis.ibanez at kitware.com>>
>  > *À :* Juliette Deniau <juliette.deniau at yahoo.fr <mailto:juliette.deniau at yahoo.fr>>
>  > *Cc :* insight-users at itk.org <mailto:insight-users at itk.org>
>  > *Envoyé le :* Mardi, 2 Décembre 2008, 21h04mn 06s
>  > *Objet :* Re: [Insight-users] active contour, snake, deformable model and level set segmentation ???
>  >
>  >
>  > Hi Juliette,
>  >
>  >
>  > Your observations are correct.
>  >
>  > If you are looking for a mesh-based implementation of a deformable model, then you should look at
>  > http://www.itk.org/Insight/Doxygen/html/classitk_1_1DeformableSimplexMesh3DBalloonForceFilter.html
>  >
>  >
>  > and its example application in
>  >
>  >    InsightApplications/DeformableModelSimplexMesh
>  >
>  >
>  > This is closer to what is known as "Snakes" in the medical field.
>  >
>  > The Level Set implementations of deformable models have all
>  > the "LevelSet" string as part of their name.
>  >
>  >
>  > They will be:
>  >
>  >
>  > $ find .. -name "*LevelSetImageFilter.h"
>  > ./itkAnisotropicFourthOrderLevelSetImageFilter..h
>  > ./itkCannySegmentationLevelSetImageFilter.h
>  > ./itkCurvesLevelSetImageFilter.h
>  > ./itkGeodesicActiveContourLevelSetImageFilter.h
>  > ./itkGeodesicActiveContourShapePriorLevelSetImageFilter.h
>  > ./itkIsotropicFourthOrderLevelSetImageFilter.h
>  > ./itkLaplacianSegmentationLevelSetImageFilter.h
>  > ./itkNarrowBandCurvesLevelSetImageFilter.h
>  > ./itkNarrowBandLevelSetImageFilter.h
>  > ./itkNarrowBandThresholdSegmentationLevelSetImageFilter.h
>  > ./itkSegmentationLevelSetImageFilter.h
>  > ./itkShapeDetectionLevelSetImageFilter.h
>  > ./itkShapePriorSegmentationLevelSetImageFilter.h
>  > ./itkThresholdSegmentationLevelSetImageFilter.h
>  > ./itkUnsharpMaskLevelSetImageFilter.h
>  > ./itkVectorThresholdSegmentationLevelSetImageFilter.h
>  >
>  >
>  >
>  > Please let us know if you have further questions,
>  >
>  >
>  >    Thanks
>  >
>  >
>  >      Luis
>  >
>  >
>  >
>  > ------------------------
>  > Juliette Deniau wrote:
>  >  > Hi All,
>  >  >
>  >  > I’m confused about what type of segmentation methods are within itk, in particular the ones related to active contour and level set?
>  >  >
>  >  > In my understanding, active contour is the same as a deformable model and the same as a snake. The levelset segmentation is another type (use a PDE to evolve the curve).
>  >  >
>  >  > In itk there are some active contours implemented with the level set framework like the CurvesLevelSetImageFilter, I’m alright?
>  >  >
>  >  > Can any one tell me what are the segmentation filters which are purely active contour (snake) and those which use level set to evolve the curve?  > Sorry if my question is too stupid!
>  >  >
>  >  > Thanks a lot
>  >  >
>  >  > Juliette
>  >  >
>  >  >
>  >  >
>  >  > ------------------------------------------------------------------------
>  >  >
>  >  > _______________________________________________
>  >  > Insight-users mailing list
>  >  > Insight-users at itk.org <mailto:Insight-users at itk.org> <mailto:Insight-users at itk.org <mailto:Insight-users at itk.org>>
>  >  > http://www..itk..org/mailman/listinfo/insight-users <http://www.itk..org/mailman/listinfo/insight-users> <http://www.itk.org/mailman/listinfo/insight-users>
>  >
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20081215/f0ad3f2f/attachment-0001.htm>