[Insight-developers] MRF and initial labeling

[Sayan Pathak]

Hi Jim/Luis,

Sorry for the delayed response. 
Regarding the labeled image not being passed to the MRF filter from a
public interface is a great observation. I think there is a room for
improvement there. I am hoping to spare some time during July/Aug to do
some necessary upgrades to the MRF filter that seems to be increasingly
gaining attention.

The labelimage is currently derived from the classifier as an automatic
initialization. This is done in line 529 in the itkMRFImageFilter.txx.
This is however not necessary if the user wants a different
initialization. I think a function that allows the user to directly set
the initial label image is a trivial fix. However, there are some issues
such as the images that the user sets must have the same number of
classes as the classifier (set by the user), this needs to be checked at
runtime, if the check fails what do we do. There is also issue of
including additional optimizers such as Graph Cuts that the Gee group is
currently working. Hopefully we can make a seamless connectivity to the
graph cuts methods with this filter.

I think these upgrades are much needed and may be collected as
requirements on the wiki page. Thougths? Opinions? I was thinking of
starting next month on improving this filter as I believe in its
potential.

Regarding the namespace issue, I am ambivalent. In early discussions it
was decided that since the filter works in the image space it came under
itk namespace much like the itkImageClassifier class hierarchy.


-Sayan
  

-----Original Message-----
From: insight-developers-bounces+sayanp=alleninstitute.org at itk.org
[mailto:insight-developers-bounces+sayanp=alleninstitute.org at itk.org] On
Behalf Of Miller, James V (Research)
Sent: Thursday, June 23, 2005 10:59 AM
To: Luis Ibanez
Cc: Insight-developers (E-mail)
Subject: RE: [Insight-developers] MRF and initial labeling

Luis, 

Thanks.  The Software Guide indicates several times that the
MRFImageFilter is in the 
statistics namespace.  It is just in the itk namespace.

Unfortunately, the example does not make use of the labeled image.  It
defines a reader for it, but it is not used anywhere.  The MRF filter is
provided 
the original intensity image and membership functions (defined in the
intensity
image space).  But the label image is not passed to the MRF filter.

Jim


-----Original Message-----
From: Luis Ibanez [mailto:luis.ibanez at kitware.com]
Sent: Thursday, June 23, 2005 12:17 PM
To: Miller, James V (Research)
Cc: Insight-developers (E-mail)
Subject: Re: [Insight-developers] MRF and initial labeling



Hi Jim,


Yes, the MRF filter accepts a labeled image as one
of its inputs.

An Example is available in


             Insight/Examples/Statistics


and it appears in the new edition of the Software Guide


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


in Section 10.4.5, in pdf-page 709-715



    Regards,


        Luis



--------------------------------
Miller, James V (Research) wrote:
> Is there a mechanism whereby I can set the initial labeling for the
MRFImageFilter? 
> 
> I have an initial classification of an image (from a process that did
not use membership functions) and I would like to refine the
classification using a set of membership functions and priors.  I'd like
to try an MRF to see what it would do to my initial classification.
> 
> Jim Miller 
> _____________________________________
> Visualization & Computer Vision
> GE Research
> Bldg. KW, Room C218B
> 1 Research Circle, Schenectady NY 12309-1027
> 
> millerjv at reserch.ge.com <mailto:millerjv at reserch.ge.com>
> (518) 387-4005, Dial Comm: 8*833-4005
> Cell: (518) 505-7065, Fax: (518) 387-6981
> 
> 
> _______________________________________________
> Insight-developers mailing list
> Insight-developers at itk.org
> http://www.itk.org/mailman/listinfo/insight-developers
> 
> 



_______________________________________________
Insight-developers mailing list
Insight-developers at itk.org
http://www.itk.org/mailman/listinfo/insight-developers