[Insight-users] surface textures?

[Luis Ibanez]

Markus,


1) You could include criteria like curvature as part of the
    metric used for evaluating the match between the two images.
    Since this is the part that you will have to write on your
    own, you can customize as you consider convenient.


2) There is a BSpline implementation in ITK,
 
http://www.itk.org/Insight/Doxygen/html/classitk_1_1BSplineDeformableTransform.html

    Probably not the one you refer from TMI.


3) The Gradient descent optimizer in ITK is pretty standard.
    There are actually two different implementations. One using the
    gradient magnitude for computing the step length, another
    using a fixed step lenght. The second is in some cases slower
    than the first one but it is more stable.


4) I don't think you can get an answer for a question like

                  "Is VTK slow ?"

    In order to get a serious answer you should rephrase the
    question as:

        "Is VTK (or ITK) slower than 'XX' for doing 'YY'  ?"

    In this context you can actually run an experiment that will
    determine unequivoaclly if the answer is 'yes' or 'no'.

    VTK and ITK are toolkits. You may eventually find filters that
    are less optimized than others. This depends on the actual
    applications intended for each filter.

5) VTK and ITK are not competing toolkit, on the contrary, they
    are intended to be complementary.

    - VTK does Visualization
    - ITK does Image Processing, Segmentation and Registration

    A full application will typically combine: ITK + VTK + GUI
    Where the GUI could be { MFC, FLTK, wxWindows, Qt..}

    In the InsightApplications module you will find many
    aplications integrating   ITK + VTK + FLTK.



   Luis



------------------------------
Markus Louw wrote:
 >
> I think youīre right that it could be posed as a registration problem,
> however I was thinking that with things like curvature information, looking
> for locations of known description (eg tip of nose with high curvature and
> high z axis component), landmarking
> would be better.
> 
> I didnīt understand this part:
> 
>>This means that the color values associated
>>with every node of the geometry will never
>>change.
> 
> We have lots of training images, each with a texture.
> 
> Also
> 
>>2) Transforms: Affine, ThinPlateSplines,
>>                VolumeSpines, ElasticBodySpline,
>>                BSpline.
> 
>>From the code, it looks like this can only be done for 2D images, not
> volumetrically, right?  (an aside question)
> Do you have anything like a BSpline Free form deformation? (Daniel Reuckert
> IEEE trans. medical imaging)
> 
> Does the Optimizer (1) work anything like the Numerical Recipes one?
> 
> Do you know anything about vtk?  Itīs slow right?
> 
> Markus
> 
>