[Insight-developers] stack registration - links between optimizer scales, image intensity, size etc

[Nicholas Tustison]

Hi Richard,

Perhaps you would be interested to know of some recent
work we did involving a significant refactoring of the registration
capabilities in ITK which includes automated scale parameter
estimation:

http://www.itk.org/Doxygen/html/classitk_1_1RegistrationParameterScalesEstimator.html
http://www.itk.org/Doxygen/html/classitk_1_1RegistrationParameterScalesFromIndexShift.html
http://www.itk.org/Doxygen/html/classitk_1_1RegistrationParameterScalesFromJacobian.html
http://www.itk.org/Doxygen/html/classitk_1_1RegistrationParameterScalesFromPhysicalShift.html
http://www.itk.org/Doxygen/html/classitk_1_1RegistrationParameterScalesFromShiftBase.html

If you would like to see a more-or-less complete tool we wrote which 
showcases the new pair-wise registration capabilities in ITKv4, I would 
recommend looking at antsRegistration.cxx (coupled with the helper 
class itkantsRegistrationHelper) available as part of our ANTs software 
repository.  

http://advants.svn.sourceforge.net/viewvc/advants/trunk/Examples/antsRegistration.cxx?revision=1521&view=markup
http://advants.svn.sourceforge.net/viewvc/advants/trunk/Examples/itkantsRegistrationHelper.h?view=markup&pathrev=1530
http://advants.svn.sourceforge.net/viewvc/advants/trunk/Examples/itkantsRegistrationHelper.hxx?revision=1525&view=markup

With respect to the problem you describe, you might also be interested
to look at a related tool also available in ANTs which uses the new ITKv4
material to perform image series registrations (e.g. for motion correction 
and image stack alignments) called antsMotionCorr.cxx

http://advants.svn.sourceforge.net/viewvc/advants/trunk/Examples/antsMotionCorr.cxx?revision=1453&view=markup

Nick



On Sep 8, 2012, at 12:16 AM, Richard Beare <richard.beare at gmail.com> wrote:

> Hi,
> It has been a long time since I played with the registration
> framework, so I've forgotten how to deal with these issues (or never
> knew).
> 
> I'm correcting a misaligned stack using 2d, pairwise, registration. I
> have an old tool based on DeformableRegistration15 in itk 3.20 and a
> newer one based on ImageRegistration7 in the itk 4.2 examples. The
> latter works well for a large proportion of the stack and suddenly
> produces a result with a very large incorrect angle. It produces that
> incorrect angle after 1 iteration. The mean squared metric is used in
> this example. Switching to Mattes MI without changing any parameters
> produces crazy results.
> 
> Are there rules of thumb for setting optimizer scales and step sizes
> based on the expected ranges of metric values and image sizes.
> 
> Thanks
> _______________________________________________
> Powered by www.kitware.com
> 
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
> 
> Kitware offers ITK Training Courses, for more information visit:
> http://kitware.com/products/protraining.php
> 
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
> 
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-developers