[Insight-users] Dense Transformations in ITK

[Luis Ibanez]

Hi Peter,

ITK already  provide registration methods based on Dense Maps.
We call them deformation fields, but are pretty much the same thing.

You may want to look at the Demons registration method, that
accepts a deformation fields as initialization, and refines it
by using optical flow.

http://www.itk.org/Insight/Doxygen/html/classitk_1_1DemonsRegistrationFilter.html
http://www.itk.org/Insight/Doxygen/html/classitk_1_1SymmetricForcesDemonsRegistrationFilter.html


You may also want to look at the LevelSet deformable registration
filter,
http://www.itk.org/Insight/Doxygen/html/classitk_1_1LevelSetMotionRegistrationFilter.html


and to the WarpImageFilter.
http://www.itk.org/Insight/Doxygen/html/classitk_1_1WarpImageFilter.html

This last one will resamples an image, by using a deformation
field as 'Transform'.  You will find examples on the use of
these filters in the directory


                Insight/Examples/Registration


BTW, if you implement a new registration algorithm, and you are
willing to contribute it to ITK, we will be happy to add it to
the toolkit.



Regards,



     Luis




----------------------------
Peter Lorenzen wrote:
> Hello,
> 
> I am considering developing an ITK-based high-dimensional image 
> registration application where the transformation involved is a dense 
> map (i.e. for n-dimensional data, an n-dimensional vector at each 
> indexed spatial location). Although ITK provides a great number of 
> transformation models, does anyone know if there are plans specialize 
> itk::Transform to implement a dense map?
> 
> To illustrate the meaning of "dense" consider:
> 
>   typedef itk::Vector< float, ImageDimension >  VectorType;
>   typedef itk::Image< VectorType, ImageDimension >  
> DenseTransformationType;
> 
> In the ITK transformation framework, we would want something along the 
> lines of itk::DenseTransform instead.
> 
> I can certainly use the itk::Image< VectorType, ImageDimension > model 
> above, but it would be nice use the existing ITK registration framework 
> (i.e., define a metric, optimizer, etc. and connect a series of existing 
> filters) instead.
> 
> Thanks,
> 
> - Peter
>