[Insight-users] point set to point set registration using deformable method (thin plate spline)

Luis Ibanez luis.ibanez at kitware.com
Thu Oct 8 08:55:32 EDT 2009


Oops...My bad,

It turned out that a computation of Jacobian
for the KernelTransforms has already been
contributed to the toolkit.

It is available in the Insight Journal paper:

http://www.insight-journal.org/browse/publication/145

"Improvements to the itk::KernelTransform and Subclasses"
Brooks R., Arbel T.
McGill Centre for Intelligent Machines, McGill University, Montreal, Canada.


I just added this paper to the list of those that we
should try to include in the toolkit by the release
of ITK 3.18:

http://www.itk.org/Wiki/ITK_Release_Schedule#Insight_Journal_papers_to_Move_for_ITK_3.18


     Regards,


           Luis


------------------------------------------------------------------------------
On Thu, Oct 8, 2009 at 8:49 AM, Luis Ibanez <luis.ibanez at kitware.com> wrote:
> Hi Tao,
>
> The itkThinPlateSplineKernelTransform doesn't provide computation of
> Jacobians. The API is there, but the method has not been implemented.
>
> You should therefore use this Transform only with optimizers that do not
> require derivatives from the Metric function.
>
> You may want to look at the Amoeba optimizer and the OnePlusOne
> evolutionary optimizers.
>
>   or....
>
> if you feel courageous today, it will be great if you give it a shot at
> implementing the GetJacobian() method in the superclass:
> itkKernelTransform, and contribute it to the toolkit.     :-)
>
>
>
>      Regards,
>
>
>            Luis
>
>
>
> --------------------------------------
> On Wed, Oct 7, 2009 at 5:14 PM, Tao Dai <uniquedai at gmail.com> wrote:
>> Hi, list:
>>
>> I am trying to register two 3D point sets with different length and without
>> known correspondence, thus I choose ICP. I mostly used the code of
>> "IterativeClosestPoint1.cxx" by replacing the transform type to
>> "itkThinPlateSplineKernelTransform". Tthe ICP went through successfully but
>> obtained NULL output parameters.
>>
>> My question is : Is the ICP working with ThinPlateSpline?  I don't see any
>> problem to use this way but can't figure it out why bad output happens. The
>> testing data I used are two sets of points, one set is 3D scaled version of
>> the other.
>>
>> I appreciate any help and inputs.
>>
>> Tao
>>
>


More information about the Insight-users mailing list