[Insight-users] Registration Metric

[Lydia Ng]

Hi Dana,

If I understand from your message you are comparing two ways of
computing the metric:

[A] You are computing the metric using the fixed image, the original
moving image and transform with parameters obtain with registration.

[B] You are computing the metric using the fixed image, the resampled
image and a transform with parameters that gives you an identity
transform.

I believe that are at least two causes of the difference in value:

[1] The metric is computed by walking over the whole fixed image. At
each point it computes (fixed_value - moving_value)^2. In [A],
moving_value is obtain by transforming the fixed image point and then
interpolating the moving image. The interpolated value is returned as
a double and the metric retains the precision when it sums things up.
On the other in [B], the moving_value has been truncated to "unsigned
char" in the resampled process.

[2] In the metric calculation, when a mapped position falls out of the
domain of the moving value, that point is not used in the calcuation.
So in [A] some number of boundary points was not used in the
calculation. However, when you resampled the image, you assigned some
value to these boundary points, so in [B] all points are used in the
calculation including those artificially assigned a "default" value.

HTH,
Lydia

On 7/12/05, Dana Paquin <dpaquin at stanford.edu> wrote:
> 
> 
> Hi,
> 
>  
> 
> I have a question about registration of images.  I am following the examples
> in the ITK User's Guide.  If I use the program ImageRegistration5 (contained
> in the Examples) to register
> BrainProtonDensitySliceBorder20.png with
> BrainProtonDensitySliceR10X13Y17.png (both included in the
> ITK distribution), I obtain the same results as in the ITK User's Guide
> (i.e. the same values for the Angle, X Translation, etc.).  This example
> uses the Mean Squares Metric, and the last metric value is 557.819.  
> However, when I compute the mean squares metric directly of my Output image
> (the final transform applied to the moving image
> BrainProtonDensitySliceR10X13Y17.png) and the fixed image
> (BrainProtonDensitySliceBorder20.png), I don't obtain
> 557.819.  Is the metric value produced by ImageRegistration5 not the same as
> the mean squares of the Output Image and the fixed image?
> 
>  
> 
> Thanks,
> 
> Dana Paquin
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
> 
>