[Insight-users] Gradient Anisotropic Diffusion

[Luis Ibanez]

Hi Mathieu,


1) Before you give up completely with option (A),...
    Did you tried running the Confidence connected
    filter ontop of this gradient magnitude image ?

    The contour doesn't have to be perfectly closed,
    a low border could be enough... depending on the
    noise level of the input image.  Just keep in mind
    that what your eyes see is not necessarily what
    the ConfidenceConnected filter sees.

    (BTW, using a viewer with intensity window capabilities
     may be a good idea, since you can better explore
     the real values on the image).

2) About the GradientAnisotropicDiffusionImageFilter:

    The reason why you don't find any mention to the
    "TimeStep" in the header of this filter is that this
    is defined in its base class, the

         itk::AnisotropicDiffusionImageFilter.

    Instead of browsing the headers you may want to
    use the documentation generated by Doxygen. If you
    want to browse code from the headers, may be you
    should consider using "SourceNavigator" which is the
    right tool for doing so. It works nicely with ITK
    and VTK. You can download it for free from RedHat.
    (a Google search will point you to the right place).


    Before applying this filter to your own image,
    please play with the example in

    Insight/Examples/Filtering/

      GradientAnisotropicDiffusionImageFilter.cxx

    using the images in Insight/Examples/Data.

    This is the code that was used for generating
    the illustrations in the SoftwareGuide.

    Once you are familiar with the parameters,
    got a cup of coffee and get ready to process
    your image.

    Note that we use very few iterations in these
    filters, however, since this is a diffusion
    process, the more iterations you apply, the
    smoother the image will become. There is a
    trade-off between smoothing and computing
    time which becomes more relevant in 3D.
    Iteration in the order of 20 to 50 are reasonable
    from the diffusion point of view, but probably
    will result in long computation times.



Please let us know if you find any difficulties.



    Thanks


      Luis



-----------------------------------------
Mathieu Malaterre wrote:
> Hi all, Hi Luis,
> 
>   As you suggested in:
> http://www.itk.org/pipermail/insight-users/2003-February/002580.html
> 
> 
> I tried first option (A). But I couldn't find no good paramaters to have 
> a closed surface.
> 
> See for differents sigma value :
> http://www.creatis.insa-lyon.fr/~malaterre/vtk/vessel_sigma5.png
> and
> http://www.creatis.insa-lyon.fr/~malaterre/vtk/vessel_sigma10.png
> 
> So I tried option (B). I tried first an edge preserving filter which is 
> to say GradientAnisotropicDiffusionImageFilter. I browse the 
> ItkSoftwareGuid.pdf and on p96, chap 5. (Filtering) you wrote:
> 
> [Typical values for the time step are 0.25 in 2D images and 0.125 in 3D 
> images.]
> 
> But if you browse now the source code of this filter there is no mention 
> to TimeStep ?
> What did I misunderstood ? What are the real parameters that affect this 
> filter ?
> 
> 
> thanks,
> mathieu
> 
> 
>