[Insight-users] convergence of curvature registration

[Torsten Rohlfing]

Hi Ruben:

In my opinion, the crucial difference between using an initial 
deformation field and a pre-resampled floating image is the boundary 
condition of the deformation. Without much theoretical insight I only 
implemented the algorithm based on Fischer & Modersitzki's papers 
(should be referenced in the source code), so I may be off here, but I 
think the problem is that while, locally, affine transformations are 
indeed in not penalized by the curvature smoother, they do violate the 
imposed boundary condition (periodic, as far as I recall). Because the 
update step is computed in a global operation using FFT, that may 
explain why the entire deformation field gets messed up.

Hope this helps.

Best,
  Torsten

-- 
--
Torsten Rohlfing, PhD          SRI International, Neuroscience Program
 Research Scientist             333 Ravenswood Ave, Menlo Park, CA 94025
  Phone: ++1 (650) 859-3379      Fax: ++1 (650) 859-2743
   torsten at synapse.sri.com        http://www.stanford.edu/~rohlfing/

     "Though this be madness, yet there is a method in't"