[Insight-users] MattesMutualInformation questions

[ND]

Hi Gabriele,

Yes, your values are correct. Please refer to the documentation of the
MattesMutualInformationImageToImageMetric class below. It says:

"Notes: 1. This class returns the negative mutual information value."

http://www.itk.org/Doxygen34/html/classitk_1_1MattesMutualInformationImageTo
ImageMetric.html#_details

My understanding is that, it multiplies the final value for the MI by -1, so
that optimizer can minimize the MI information instead of maximization. This
is important since some optimizers are only able to minimize, but not
maximize. So the actual MI value in your last step (0.26143) is greater than
the first step (0.190391): you just have to ignore the (-) sign.

Cheers,
Nuwan


-----Original Message-----
From: insight-users-bounces+nuwan.maillists=gmail.com at itk.org
[mailto:insight-users-bounces+nuwan.maillists=gmail.com at itk.org] On Behalf
Of gabri
Sent: November 27, 2007 12:19 PM
To: Insight User
Subject: [Insight-users] MattesMutualInformation questions

Hi all,
I would like to understand how to fine tune an Affine3DRigid
Registration using MattesMutualInformationImageToImageMetric.
The simple question is :
I suppose that in theory the MI final value would be  maximized, for a
correct registration result.
But in my code the final value is minor than the first step of the
itaration process.
first ->
0   -0.190391   [0.995462, -0.00157196, 0.00313902, 0.000640376,
1.00341, -0.00225071, 0.000649128, -0.00112494, 0.997894, -3.2308,
-4.89763, -18.4037] AffineAngle: 0.0634048


last ->
80   -0.26143   [0.993057, -0.00753589, 0.0122628, 0.000674591, 1.00105,
0.00183965, 0.00413924, -0.005144, 0.991511, -3.08394, -4.94793,
-18.5664] AffineAngle: 0.234475

why my values are minor than zero? is this correct ?
The images that my code produce are fine coregistered but i could not
understand (search in the itk handbook and also in mailing list's
previous post) why the values are negative...

Best reguards ,
Gabriele