[Insight-users] Why is the Normalized Correlation Metric inverted?

[Jonas Beuke]

Hi all,

I'm comparing images by using itkNormalizedCorrelationImageToImageMetric 
as a similarity metric. As far as I can see it implements exactly the 
Pearson product-moment correlation coefficient except for the line 117 
"const RealType denom = -1.0 * vcl_sqrt(sff * smm);". This line, as far 
as I understand it, inverts the correlation so -1 is a perfect positive 
correlation between the two images. What is the purpose of this?

Greetings,
Jonas