[Insight-users] Re: About the Gaussian Mixture Model estimation

[Luis Ibanez]

Hi Jimmy


You will find multiple examples on the use of the
ITK Gaussian Mixture Modeling:


1)  The ITK Software Guide

         http://www.itk.org/ItkSoftwareGuide.pdf

     Section 10.3.3, pdf-page 475.



2)  The source code of the Validation Study on
     tissue classification, available at


        InsightApplications/
                IBSRValidation/
                      IBSRClassification/
                                      Code



3)  The report of the validation study in (2) that
     is available at


          InsightDocuments/
                    Validation/
                          TissueClassification

http://www.itk.org/cgi-bin/viewcvs.cgi/Validation/TissueClassification/?root=InsightDocuments

     You probably want to read the report in PDF format

http://www.itk.org/cgi-bin/viewcvs.cgi/*checkout*/Validation/TissueClassification/TissueClassificationValidationStudy.pdf?rev=1.1&root=InsightDocuments




4)  The examples recently added to ITK for the use
     of the K-Means classifier and the Markov Random
     Fields. They are not directly related to the GMM
     but make use of a similar family of classes.


              Insight/Examples/Segmentation/
                  ScalarImageMarkovRandomField1.cxx
                  ScalarImageKmeansClassifier.cxx




The Classification framework in ITK is extremly flexible.
You control almost every piece of the process. You pass
the Membership functions to the classifiers and to the
estimators. So you have the freedom to redefine any of
the components and make your own combinations.


Please let us know if you have further questions,



    Thanks



      Luis



---------------------
Jimmy Wong wrote:

> Dear Luis,
> 
> Thanks for help me with the previous question.
> 
> Now, I got another one. I know I can use GaussianMixtureModelComponent & 
> ExpectationMaximizationMixtureModelEstimator to estimate the parameters 
> for the GMM. So, can I use it to classify the samples (or do the 
> clustering) ? Any example for it?
> 
> Can I fix the proportion (weight) of each component in 
> ExpectationMaximizationMixtureModelEstimator and estimate the new 
> gaussian parameters?
> 
> I want to use it in the registration framework and hope can get the 
> membership function point from this estimator. But it seems that GMM 
> estimator does not provide any function for return or obtain the 
> membershipfunction. Let me know if I am wrong.
> 
> Then, in this case, what can I do?
> 
> Thanks.
> 
> 
> 
>