[Insight-users] Normalized Mutual Information Metric

[Raghavendra Chandrashekara]

Thomas Boettger wrote:
> Does anyone know "how usable" the class 
> NormalizedMutualInformationMetric is? What are the Derivative StepLength 
> and Scale for? And which parameters need to be set. I couldn't find an 
> example. Is it still to early to use this metric?
> 
> And why isn't there a binning parameter? Does the histogram size 
> determine the number of gray value classes and has the same affect as 
> binning?
> 
> Thomas
> 
> 
> 
Hi Thomas,

Sorry about the delay in replying. I was one of the people who helped to 
  write the code for this metric. If you still need some help then 
please let me know and I will try to answer your questions. As for the 
questions you asked previously:

(1) What is the derivative step length?

The derivative is calculated using finite differences. The derivative 
step length gives the step length for each transformation parameter when 
the derivative is being calculated.

(2) What is the scale for?

Since the transformation parameters are not always distances (e.g. they 
could be rotation angles), the derivative step length scales allow you 
to specify a scaling factor for the step length corresponding to a 
particular transformation parameter. The scaling factor is used when the 
derivative is being calculated to scale the step length for that 
transformation parameter.

(3) And why isn't there a binning parameter? Does the histogram size 
determine the number of gray value classes and has the same affect as 
binning?

Yes that's right the histogram size determines the number of gray value 
classes.

There is an example on how to use the metric in:

Testing/Code/Algorithms/itkNormalizedMutualInformationHistogramImageToImageMetricTest.cxx

Please let me know if you need any more help.

Thanks,

Raghavendra