[Insight-users] normalized historgram MI metric

[Christos Panagiotou]

hi guys

i am trying to use the normalized histogram mutual information image 2 
image metric

to do this i am trying to get an idea of how the class is used by the

Testing/Code/Algorithms/itkNormalizedMutualInformationHistogramImageToImageMetricTest.cxx

however i ve got some questions

i want to test the registration results using this class (my data is 2 volumes (different modalities) around 256x256x100)

1. nBins what would be a good choice of the number of histogram bins and how do we calculate this...
2. about derivative step length scales
the scales for an affine transformation would again affect the last 3 components of the matrix (10, 11, 12)
and are calculated using: scales = 1.0 / 10 * sqrt( x^2+y+2+z^2)    ? i am asking because a derivative is involved
3. the histogram is always compromised of 2 components with the same nBins?
4. can i use this class with the itkMultiResolutionImageRegistrationMethod in the following manner?
  m_Registration -> setMetric( normalized mutual information histogram image2image metric)
  and then initialize the registration?
5. any additional suggestions of how to use the metric?



is this metric slower than the MIImage2Image metric ?

thanks
christos