[Insight-users] Affine+Mattes 2D registration problem

[Luis Ibanez]

Hi Manuel,


1) Please post to the list the output values of Metric
    and Transform parameters produced at every iteration
    of the registration.

2) You may want to start by using a very large number of
    samples.

3) Could you please post a screenshot of the outcome ?
    Your statement that the "results are not good" is
    not informative enough to advise you.

    For example, it is common for Mutual information to
    find minima in positions of the image that will overlap
    large regions to large regions. (even if they are not
    related).


4) Note that if your motivation for changing the metric
    was to reduce the computation time, by taking advantage
    of the pixel sampling use by MattesMutualInformation,

    You can now have the same sampling functionality in the
    MeanSquaresMetric if you turn ON the CMake option:

           ITK_USE_OPTIMIZED_REGISTRATION_METHODS

    This may be a better way to proceed, given that
    you already have a successful registration using
    that metric..



   Regards,



      Luis



----------------------------------
Manuel David Tabas Calle wrote:
> Hi!
>  
> I have a problem using an Affine transform with Mattes Mutual metric.
>  
> When I used an Affine transform, with MeanSquaresImageToImageMetric, my 
> registration works perfectly, but it is really slow because my images 
> are big (it last for 45 minutes). My images are between 2000x2000 and 
> 4000x4000 pixels, with spacig of (1,1,1).
>  
> So i decided to use MattesMutualInformationImageToImageMetric, that is a 
> faster metric. But the results are not good.
>  
>  
> In both, I used the followin parameters:
>  
> optimizerScales[0] = 1.0;
> optimizerScales[1] = 1.0;
> optimizerScales[2] = 1.0;
> optimizerScales[3] = 1.0;
> optimizerScales[4] = 1.0/1000.0;
> optimizerScales[5] = 1.0/1000.0;
>  
> steplength = 0.1
>  
> maxNumberOfIterations = 300;
>  
> minimumStepLength = 0.0001;
>  
> and
>  
> optimizer->MinimizeOn();
>  
> For the Mattes Mutual metric, I tried a lot of combinations of histogram 
> bins and spatial samples, starting with 50 bins and a 20% of the pixels 
> as spatial samples. But no matter what i try, the results are not good.
>  
> In both programs, i use LinearInterpolateImageFunction and 
> RegularStepGradientDescentOptimizer.
>  
> How can I tune the parameters of the program with MattesMutual to obtain 
> the same results???
>  
> Thank you all!!!
> 
> ------------------------------------------------------------------------
> Charlas más divertidas con el nuevo Windows Live Messenger 
> <http://download.live.com>
> 
> 
> ------------------------------------------------------------------------
> 
> _____________________________________
> Powered by www.kitware.com
> 
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
> 
> Please keep messages on-topic and check the ITK FAQ at: http://www.itk.org/Wiki/ITK_FAQ
> 
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users