[Insight-users] Mutual Information Metric

[soosho]

Dear Luis

This is Mehdi Esteghamat. I am using ITK for real-time (RT) cardiac
image registration. I employed mutual information metric. I timed
registration period that should be less than 80ms in my RT circumstance. 


But something is wired. I could see that sometime the registration take
places in 30ms (24 iterations of regular step gradient descent
optimizer) and some other times it takes 270ms. Added to this running
time is a constant value for each compilation of code. I mean for some
compilation the running time of RT registration process is 30ms and for
some other compilations it is 270ms.

I guess that some important code variables settle in cash in some
compilations but in other compilations it remains in memory and produces
many page faults. 

It is noteworthy that number of samples is 450 for both cases. Also, the
hardware setting is :

Linux localhost 2.6.17-5mdv #1 SMP Wed Sep 13 14:32:31 EDT 2006 i686
Intel(R) Pentium(R) 4 CPU 3.40GHz GNU/Linux

RAM: 1 GB dual channel

I employed mattes implementation of mutual information
(itk::MattesMutualInformationImageToImageMetric).

What I like to know what could be the probable reason for such behavior
from ITK?

Best Regards,
Soosho
-- 
  soosho
  esteghamat at fastmail.fm

-- 
http://www.fastmail.fm - The way an email service should be