Hi Benny,
> You can re-smooth a smoothed image to increase the amount of smoothness,
> although you'd need to calculate the effective smoothing, which isn't
> entirely trivial.
For Gaussian smoothing, I think variances theoretically add, so FWHMs
or sigmas for ip add in quadrature.
I've just had a very quick check of this in practice, and it doesn't
perfectly hold around the edges of the image volume, due to boundary
effects, but over the brain, smoothing with sigma=3 then sigma=4 is
almost indistinguishable from smoothing with sigma=5. E.g. using
$FSLDIR/etc/standard/avg152T1 and its brain_mask, avwstats robust
range (-r) of the difference image is (-0.000571,0.000968).
For non-Gaussian smoothing, especially non-linear methods such as
median filtering, I think it would be very difficult (perhaps
impossible) to determine effective smoothing from multiple passes.
I hope that helps,
Ged.
|