[Insight-users] Re: itk Gradient Descent Optimizer : Mixture modeling ?

[Luis Ibanez]

Hi digvijay,

What you are doing looks like a good fit
for Mixture Modeling. That is, fitting a
set of Gaussians to an histogram.

You may want to take a look at the Statistics
package and Classification Framework in ITK.
This can be found at

    Insight/Code/Numerics/Statistics

and

    Insight/Code/Algorithms


The following classes could be good start

GoodnessOfFitMixtureModelCostFunction
http://www.itk.org/Insight/Doxygen/html/classitk_1_1Statistics_1_1GoodnessOfFitMixtureModelCostFunction.html

MixtureModelEstimator
http://www.itk.org/Insight/Doxygen/html/classitk_1_1Statistics_1_1ExpectationMaximizationMixtureModelEstimator.html

GaussianMixtureModelComponent
http://www.itk.org/Insight/Doxygen/html/classitk_1_1Statistics_1_1GaussianMixtureModelComponent.html

SampleClassifier
http://www.itk.org/Insight/Doxygen/html/classitk_1_1Statistics_1_1SampleClassifier.html

TableLookupSampleClassifier
http://www.itk.org/Insight/Doxygen/html/classitk_1_1Statistics_1_1TableLookupSampleClassifier.html




Please let us know if you have further questions,


   Thanks


    Luis


=======================================================

digvijay singh wrote:
> hi luis!!
> i get the emphasis on const......and the optimizer
> does not implement constraints....bad news for me :-),
> now i am trying to read in an image compute its
> histogram and model it as a sum of gaussian functions.
> The model gets  good enough values for the
> initialization from histogram computation as  one of
> the gaussians is predominant over the rest. I was
> trying to minimize the error in the fit starting with
> a single gaussian but i need to impose constraints on
> the way the optimization progresses simply because i
> wish to compute it for the predominant gaussian first.
> That's my side of story....thanks for your help
> ciao
> digvijay
> 
> 
>