[Insight-developers] mattes optical flow

[Mathieu De Craene]

Hi Daniel

Sorry for the delay but in the meantime, I worked on another
implementation giving the possibility to control the width of the kernel
function used for estimating the joint probability density.

As you requested, I wrote a short documentation with a reference to some
papers. You can get it as this URL : (pdf, html and tex files)
     http://www.tele.ucl.ac.be/~decraene/index.php?menu=code_miflow

My source code can be downloaded from
          http://euterpe.tele.ucl.ac.be/Waleo2/miflow/src.tgz
There is a test copied from the Demon's test.

My next step is to constrain the bijectivity of the displacement field
following the scheme described by Thirion. I found this mail on the
users mailing list :
http://public.kitware.com/pipermail/insight-users/2004-November/011241.html but nothing more recent ;-/ 

Is bijectivity intended to be implemented soon in ITK ?

Thanks for your comments !!

Mathieu.


Le mardi 19 avril 2005 à 10:18 -0400, Blezek, Daniel J (Research) a
écrit :
> Mathieu,
> 
>   Interesting!  I'm happy to commit your code to ITK, if you will add documentation (including a reference to the paper), and a test of the code (just modify the Demon's test).
> 
> Thanks,
> -dan
> 
> -----Original Message-----
> From: insight-developers-bounces at itk.org
> [mailto:insight-developers-bounces at itk.org]On Behalf Of Mathieu De
> Craene
> Sent: Tuesday, April 19, 2005 9:37 AM
> To: insight-developers at www.itk.org
> Subject: [Insight-developers] mattes optical flow
> 
> 
> Hi Itk colleagues !
> 
> I have put on this directory
> 
>              http://www.tele.ucl.ac.be/~decraene/src-miof/
> 
> a copy of some classes I wrote to use the derivative as implemented in
> the itkMattesMutualInformationMetric in a optical flow registration
> framework. Feel free to commit these classes in the Insight tree if you
> find them useful.
> 
> I have seen that another class (itkMIRegistrationFunction) is doing the
> same job for Viola's implementation. Basically, any metric derivative
> which can be written like
>       Deriv = \Sum_samples TransfoJacobian(sample) * flow(sample)
> can be seen as the sum of the projections of a flow vector
> ( flow(sample) ) on the transformation jacobian at this point.
> 
> Could we imagine some day to have an intersection between the two
> registration families in ITK ? Having some metrics which could be
> optimized with parametric transformations using the GetDerivative()
> method OR in a dense deformation field estimation at every voxel like in
> Optical Flow ?
> 
> Thanks for sharing your thoughts ;-)
> 
> 
> Math.
> 
>