[Insight-users] can itkMRFImageFilter discard the intensity information

[Baoyun Li]

Hi, Luis:

I have look the following code in   Insight/Examples/Statistics/ ScalarImageMarkovRandomField1.cxx


typedef itk::Statistics::DistanceToCentroidMembershipFunction<
ArrayPixelType >
MembershipFunctionType;
typedef MembershipFunctionType::Pointer MembershipFunctionPointer;
double meanDistance = 0;
vnl_vector<double> centroid(1);
for( unsigned int i=0; i < numberOfClasses; i++ )
{
MembershipFunctionPointer membershipFunction =
MembershipFunctionType::New();
centroid[0] = atof( argv[i+numberOfArgumentsBeforeMeans] );
membershipFunction->SetCentroid( centroid );
classifier->AddMembershipFunction( membershipFunction );
meanDistance += static_cast< double > (centroid[0]);
 
}
 
Certainly, the program need user to input the centroid which I guess is the mean intensity value of each cluster? Am I right?
 
My qustion is how accurate the program relies on the input mean. Since I already have reference labeled image, it is not a big deal to calculate the mean.
Can you furthe tell me whehter I can get the similar effect by using itk EM clustering and itkMRFImageFilter with the method by Zhang et al. (many again is a naive question)
 expectation-maximization. 
 
Thanks
 
Baoyun.



________________________________
From: Luis Ibanez <luis.ibanez at kitware.com>
To: Baoyun Li <baoyun_li123 at yahoo.com>
Cc: insight-users at itk.org
Sent: Saturday, March 21, 2009 12:35:48 PM
Subject: Re: can itkMRFImageFilter discard the intensity information


Hi Baoyun,


Please take a look at the example:

  Insight/Examples/Statistics/
      ScalarImageMarkovRandomField1.cxx

You may find it to be a useful guide on how to use the MRF filter.

I'm not sure what you mean by "using the intensity information".
Could you please explain this in more detail ?

A piece of source code will be great....


    Thanks


        Luis



------------------
Baoyun Li wrote:
> Dear All:
>  After going through the documents for itkMRFImageFilter, I have some doubts about using the intensity information.
>  I understand the filter using Gassian model to relabel the segmeantion image. But why?
>  In otherwords, if I have already got ok segmeantion result based EM Gaussian Mixuture model, thus I only need hMRF to take care of the continous of the labled segmetation. Seem reclassify the image based on Gassuian model is redundent or make the segmentation worse if I blieved my initial segmentation.
>  If there anyway to ignore using the intensity information.
>  Or I am fundmentally wrong, say that using intensity information certainly can improve the performance.
>  Can somebody teach me?
>  Best regards
>  Baoyun
> 
Zhang, Y., M. Brady, and S. Smith. Segmentation of brain MR images through a hidden Markov random field model and theIEEE Trans. Med. Imaging 20:45–57, 2001


      
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20090323/f45a0f9a/attachment.htm>