[ITK-dev] [ITK] Setting minimum and maximum values for parameters in Amodea optimizer

[Yaniv, Ziv Rafael (NIH/NLM/LHC) [C]]

Hello Marie,

The Amoeba (downhill simplex) method is an unconstrained optimization method and therefore will allow for any real value. An artificial way that may work is that your similarity measure return itk::NumericTraits<float>::max()  when one of the parameters is outside the valid range. I am assuming here you are minimizing the function and using float. This is breaking the heuristic underlying the downhill simplex method, but it may get you what you want.

If you are using the ITK3.2 registration framework (still part of the 4.x releases) there are two optimizers that accommodate simple bound constraints:

http://www.itk.org/Doxygen/html/classitk_1_1LBFGSBOptimizer.html
http://www.itk.org/Doxygen/html/classitk_1_1ParticleSwarmOptimizer.html

If using the ITK4 registration framework there is:
http://www.itk.org/Doxygen/html/classitk_1_1LBFGSBOptimizerv4.html

Please take care to read the correct documentation when selecting an optimizer. Those that are part of the v4 registration framework have “v4” appended to their name (see the LBFGSB optimizers above).

          hope this helps
                     Ziv
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-developers/attachments/20150717/9569808d/attachment.html>