The coregistrion in SPM is treated as an optimisation problem, and the
optimisation algorithm used searches for a minimum. It therefore seemed
easiest to minimise the negative of the mutual information, rather than
maximise the mutual information.
The function differs from that of mutual information because you are using
normalised mutual information (and not mutual information) to do the
% Studholme, Hill & Hawkes (1998).
% "A normalized entropy measure of 3-D medical image alignment".
% in Proc. Medical Imaging 1998, vol. 3338, San Diego, CA, pp. 132-143.
Thanks for the reply. But Isn't the mutual information a non-negative value?
When I co-register a image to itself using Normalized mutual information
method, I got the number in the last column of last line: -1.218. when I
co-reg a fMRI to anatomical image, I got -1.0698. And the negative value is
inconsistent with the definition of mutual information.
I would expect to see the mutual information for registering an image to
itself has biggest value than registering to a different image. By the way
I'm using latest SPM8.