[Insight-users] GLCM Histogram computation speed

[Neil Panjwani]

I'm using OTB to compute Haralick texture features locally.

http://www.orfeo-toolbox.org/doxygen/classotb_1_1ScalarImageToTexturesFilter.html

However, the filter is extremely slow (even in release mode). I've only run
it on one GLCM offset at a time, and I will eventually need to run it over
several and average the result. I believe the step which computes the GLCM
histogram for each voxel is slowing it down.

My question is with the use of a specific ITK filter:
MaskedScalarImageToGreyLevelCooccurrenceMatrixGenerator

http://www.itk.org/Doxygen318/html/classitk_1_1Statistics_1_1MaskedScalarImageToGreyLevelCooccurrenceMatrixGenerator.html

I read a tutorial on texture and GLCM here:

http://www.fp.ucalgary.ca/mhallbey/properties.htm

which says the GLCM should be compressed first or I risk having too many 0s
in my matrix.

Is this true? Do I need to compress my data? If so, what's the best way of
going about it?

How is this related to the SetNumberOfBinsPerAxis() method, i.e. if the
number of bins is fixed (default: 256) will the computation be faster if I
compress my data?

Note I also do have to set the input minimum and maximum, which affects the
computation...

Thanks in advance!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20110526/00afc31a/attachment.htm>