[Insight-users] Normalizing GLCM feature values

[Zachary Pincus]

The theoretical max and min of the features could be calculated  
according to the definition of the GLCM (see the documentation for  
the class that generates GLCMs, e.g.  http://www.itk.org/Doxygen/html/ 
classitk_1_1Statistics_1_1ScalarImageToGreyLevelCooccurrenceMatrixGenera 
tor.html ) and the definition of the features themselves (http:// 
www.itk.org/Doxygen/html/ 
classitk_1_1Statistics_1_1GreyLevelCooccurrenceMatrixTextureCoefficients 
Calculator.html ). Note that the theoretical max and min will depend on:
(1) The pixel type
(2) The number of bins in the GLCM
(3) Whether the GLCM is normalized

and of course, the actual feature values you get may not have a  
dynamic range of anywhere *near* the theoretical range, depending on  
the images you use, so normalizing the theoretical range to 0-255  
might well not buy you very much at all.

A potentially better approach would simply be to get a set of images  
of the same type that you plan to use, and to determine the mean and  
standard deviation of the feature values for these images. Then use  
these values to standardize future feature values. Note that using  
means and standard deviations will also be less brittle in this case  
than forcing values to be within 0-255, because it allows for the  
possibility that atypical images will have feature values beyond  
those you observed in the training set.

Without knowing the specifics of your need, I can't say too much, but  
I do think that taking the empirical approach to determining the  
dynamic range of feature values for images that interest you is best.  
If you really need to know the theoretical max and min of the feature  
values, the documentation linked above should be sufficient to  
calculate it with a half hour's time or so.

Zach Pincus

Program in Biomedical Informatics and Department of Biochemistry
Stanford University School of Medicine


On Sep 29, 2005, at 12:15 AM, <Aster.Wardhani at csiro.au>  
<Aster.Wardhani at csiro.au> wrote:


> Hi,
>
> I’d like to normalize the feature values obtained from using the  
> ScalarImageTextureCalculator so that they have the same range  
> (0-255). Does anyone know the max value for each feature is?
>
> Thanks,
>
> Aster.
>
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
>
>