[Insight-users] Normalized Mutual Information

[sadegh]

Hello Maite

Well, based on my experience, there is a quite difference between
"Calculating MI STATISTICS between two images" for only an evaluation and
"Calculating MI as a similarity metric in a procedure of image
registration". You have to see which you're looking for. The codes in
"MutualInformationImageToImageFilter.cxx" seems wired to me and that is
because the same sources have been used for both input images.

ImageType::Pointer fixedImage = randomImageSource1->GetOutput();
ImageType::Pointer movingImage = randomImageSource1->GetOutput();

Anyway, why don't you simply deploy
"itkMattesMutualInformationImageToImageMetric.cxx" and use its "GetValue"
function in order to calculate MI between the images. However there is
another great Submission code in Insight Journal 
http://www.insight-journal.org/browse/publication/166
which has a useful code to calculate the MI and NMI using "JointImageFilte".
As you see in the code description, NMI is not a big deal! "Normalized
Mutual Information is calculated as, value of Mutual Information gets
divided by the mean entropy of the input images."

Hope this short descp. helps.

Sadegh

MRUIZ wrote:
> 
> Hi, 
> 
> I am using the MutualInformationImageToImageFilter.cxx source code of the
> examples to calculate the mutual information of two images. This program
> calculates two different mutual information parameters. Could anyone tell
> me what does Normalized Mutual Information 1 refers to? I know the
> definition of mutual information as the ratio between the sum of
> entropynform of the images and the joint entropy. 
> But this program gives me two Normalized Mutual information parameters... 
> 
> Thanks in advance!
> 
> Maite  
> 
-- 
View this message in context: http://old.nabble.com/Normalized-Mutual-Information-tp33544831p33544833.html
Sent from the ITK - Users mailing list archive at Nabble.com.