[Insight-users] Model-based-Registration strange results?!?

[Luis Ibanez]

Hi Gunther,

The output of the example is encoding the parameters
of the Euler2DTransform as :

    [ angle(radians), translation X (mm), translation Y (mm) ]

Why do you say that the output differs from the figure ?

The figure plotting the path of the optimization is
illustrated in page 339, Figure 8.42. The optimizer
starts at

     angle        = 0.2 rad
     translationX = 7.0 mm
     translationY = 6.0 mm

and finishes at

     angle        = 0.002 rad
     translationX = 0.53  mm
     translationY = 0.28  mm

which is pretty close to [0,0,0]. (the theoretical solution).


Keep in mind that this example is using an evolutionary
algorithm as optimizer, which is evolving a random population
of samples in the parametric space.


----

If you want to get advice in the version that you modified
from this example,...
well...
it will be very nice on your part to post the code to the list   :-)



Thanks,


   Luis



------------------
Gunther Sudra wrote:

> Hi,
> 
> when using Model-Based-Registration to detect simple objects I got sometimes very good and sometimes very strange results. Therefore I took a closer look at the example and have the following question:
> 
> on page 339 of ITKSoftwareGuide are two figures about the development of angle and x,y-translation during optimization.
> 
> when I run the examples I get the following (final) result [0.00279581, 0.530624, 0.280785]
> 
> I interpret this like
> 
> angle: 0.00279581 
> translation x: 0.530624
> translation y: 0.280785
> 
> but these results differ from the plot in the figures. How can this be?
> 
> 
> I also tried to detect an artificial ellipse with an ellipse-model. In some cases (especially when there are large differences in orientation) results were very poor. Even an large increase of number of iterations (>2000) and number of initializations (>2000) didn't lead to better results. 
> 
> All I did was  simplifiying the example, so that I have now one ellipse-model to detect the same rotated and translated ellipse in a 2D-Picture.
> 
> Am I doing something completely wrong?
> 
> Best regards
> Gunther
> 
> 
> 
>