[Insight-users] Parameters of ICP algorithm

[Luis Ibanez]

Hi Jonathan,

The ICP algorithm in ITK follows the same
principle as the one in VTK. That is, to
progressively minimize the sum of squared
distances between two sets of points.

However, the implementation is quite different since
in ITK it takes advantage of the Registration
Framework components.

In particular, it uses the Levenberg-Marquardt
optimizer instead of a gradient descent one. This
allows to take advantage of the local information
provided by point-to-point distances instead of
limiting the optimization to the search of the minimum
value for the total sum of squared distances. In this
way rotations are resolved faster that when only the
total sum of distances is computed.

The open framework for registration in ITK allows
you to do things like experimenting with your own
variations of the point-to-point metric, and still
reuse the rest of the registration components.


You may want to look at the source code example in


   Insight/Examples/Registration/
              IterativeClosestPoint1.cxx
              IterativeClosestPoint2.cxx
              IterativeClosestPoint3.cxx



For an overview of the Registration Framework in ITK,
please look at the tutorials


     http://www.itk.org/HTML/Tutorials.htm


In particular to


http://www.itk.org/CourseWare/Training/RegistrationMethodsOverview.pdf


You should also look at the SoftwareGuide


    http://www.itk.org/ItkSoftwareGuide.pdf


and its chapter on registration


    Chapter 8, pdf-page 241 to 340.




Regards,


    Luis







------------------------
Jonathan Bailleul wrote:


> A newbie question: it is not in the current online documentation?
> Does it differ from its vtk brother?
> 
> Yes, I do some vtk and just have a look of what itk is...
>