[Insight-users] EM segmentation with MRF

[Luis Ibanez]

Hi Jimmy,


A) You will find a full example on how to use
    the ITK Statistical Classification classes
    in the directory


     InsightApplications/
              IBSRValidation/
                   IBSRClassification/


    You will find there examples on how to perform
    Gaussian Mixture Modeling, K-Means, and how
    to combine them with Markov Random Fields.

    This is the code developed by Insightful for
    a validation of the statistical methods.
    A report on this validation study is avalable
    in the InsightDocuments CVS checkout.



B) You will find an example on how to perform
    KMeans, and another on how to use the Markov
    Random Field in the directory:


    Insight/Examples/Statistics/
       ExpectationMaximizationMixtureModelEstimator.cxx
       ScalarImageKmeansModelEstimator.cxx
       ScalarImageMarkovRandomField1.cxx
       ScalarImageKmeansClassifier.cxx


    The container used for the class labels is an itk::Image,
    whose pixel type is usually an integer type (e.g. char,
    int, short...etc). You will see how this is done in the
    examples in Insight/Examples/Statistics.

    These examples were *recently* added, so you will *NOT*
    find them in the ITK release 1.8. You *must* use a recent
    CVS checkout if you want to take a look at these examples.




C) For instructions on how to read values from ITK images,
    please read the ITK Software Guide

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


    in particular Chapter 4, Section 4.1, pdf-page 61-73.




D) NOTE that labels are usually produced with values
    0,1,2,3,4.... etc, and therefore, when you attempt
    to look at these images of labels using naive viewers,
    you may find that the image "seems to be empty".
    If you plan to use the labeled image for visualization
    you will have to rescale the intensities in the image
    in order to better use the dynamic range.

    For example, note that the filter:
    ScalarKmeansImageFilter
http://www.itk.org/Insight/Doxygen/html/classitk_1_1ScalarImageKmeansImageFilter.html

    and its methods

            GetUseNonContiguousLabels
            SetUseNonContiguousLabels



E) You will also find a K-Means ITK plugin that was
    recently added to  VolView.

    The source code is available in

           InsightApplications/
                    VolviewPlugins


    The free binary version of VolView can be downloaded
    from

      http://www.kitware.com/products/volview.html



F) If you are writing new classes for Statistical
    classification, you *must* read the ITK Software
    Guide chapter on Statistics,

    This is Chapter 10, in pdf-pages 421-478.





Regards,



       Luis




--------------------------------------------------------
Jimmy Wong wrote:
> Hi, dear insight-users,
> 
> If I want to simulate the image segmentation by using EM algorithm, 
> Gaussian Mixture Model and Markov Random Filed, is any template 
> available for this, no matter it is test version or not?
> 
> Another thing is, if I segment the image via a particular method, then 
> what I have are the class labels? What kind of container I should use to 
> store the labels if I want to use these labels in ITK toolkit for 
> displaying segmented image, or constructing a mask or something like that?
> 
> You know, the particular algorithm or function is written by me, but I 
> want it to be  compatible with ITK.
> 
> In a word, the second question is: for segmentation, we mainly treat 
> with the intensity value? How can I extract the intensity value from the 
> ImageReader or Image type data? And how can I recover the segmentation 
> information or reuse the labels in ITK again? What kind of data 
> structures should I use for the intensity value and class labels.
> 
> Thank you very much.
> 
> Zhimin
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>