[Insight-developers] Memory efficient gaussian mixture models

[Richard Beare]

Hi,

Is there a simple way of adapting
ExpectationMaximizationMixtureModelEstimator so that histogram
versions of the samples can be used? I'm hoping to compute parameters
of mixture models for a reasonable number of MR scans combined, and
adapting the current implementation to a histogram or map based
approach seems an obvious way to go. This will probably require a new
version of  itkGaussianMixtureModelComponent.h and
itkWeightedMeanCalculator etc, but I was wondering whether there are
any short cuts I might have missed.

Thanks