[Insight-users] 3D rigid image registration

[Luis Ibanez]

Hi Ghassan,

The advantage in the way MutualInformation
is computed is that it uses only 50 sample
pixels to compute the metric value.

This pixels are selected randomly from the
image using the itk::RandomImageIterator.

The other metrics are computing the value
using *all* the pixels in the FixedRegion
defined by the user. This is typically
much bigger than 50 pixels....

We could imaging to use the RandomIterator
too in the other metrics and estimate
their values by sampling... It will require
some tests to find the sweet spot between
computing time and the qualitiy of the
estimation. (e.g. how many sample pixels to
take)

Curiously this will require minimum changes
on the metrics...


   Luis


============================================
Ghassan Hamarneh wrote:
> Hi Luis,
> 
>  --- Luis Ibanez <luis.ibanez@kitware.com> wrote: 
> 
>>    MutualInformation should still be capable of
>>    registering same modality images anyways and
>>    compared to the other metrics will be much
>>    shorter computing times.
>>
> 
> Is that true?... I was under the impression that MI is generally more
> computationally intensive than other similarity mesures.  Comments?
> 
> /Ghassan
> 
> 
> 
> 
> ______________________________________________________________________ 
> Post your free ad now! http://personals.yahoo.ca
> 
>