> as John keeps pointing out, he considers the normalization through
> segmentation to be superior to the standard normalization. I wonder now,
> is this true for the affine part, too, or are they derived using the
> same algorithm?
The affine part is done differently. It uses a mutual information objective
function, rather than a combined segmentation model. Soon, there will be an
update for SPM5 that will make the affine part a bit more robust and
hopefully more accurate.
>
> More concrete, is it at all possible to only use the affine part of the
> segmentation normalization parameters to write native-space images, and
> if so, does it make sense?
You could try a very low frequency cutoff for the nonlinear warping, which
will just model the deformations by three translation parameters, but I would
expect the segmentation to be more accurate if more nonlinear deformations
were to be modelled.
Best regards,
-John
|