[Insight-users] NormalizedCorrelationImageFilter producing infinity?

[David Doria]

I am trying to make a demo of NormalizedCorrelationImageFilter with a mask:

http://www.itk.org/Wiki/ITK/Examples/Broken/Images/NormalizedCorrelationImageFilterMasked

Without a mask, if I create two images that are a translation of one
another, the resulting maximum correlation value is 1 and the offset
is exactly the index of the maximum (minus the patch radius). However,
even before I got to trying the masked correlation, I ran into a
problem. I created the same translated image, but this time set the
value under the soon-to-be masked region to 122 (the images were
previously binary with values 0 and 255). I expected the resulting
correlation value to be some value less than 1, and for the index of
the maximum to be essentially arbitrary. However, the result I get is
infinity:

Maximum location fixed: [25, -25]
Maximum value: inf

You can reproduce this by commenting/uncommenting this line:

FillRegion(image2, maskRegion, 122);

Can anyone explain why the correlation value would ever be infinity?

Thanks,

David