Hi Mark,
Thanks for you quick respons, one question remains: is it possible to reduce
the effect with any of the sinc-options, such as -sincwidth or -sincwindow ?
Thanks,
Erik-Jan
On Friday 11 April 2003 02:51 pm, you wrote:
> Hi Erik,
>
> Probably not a bug, as unfortunately this is a well-known phenomenon
> with sinc
> interpolation. It is known as Gibbs ringing and happens when it tries
> to interpolate
> a sharp boundary, causing slight overshoots and undershoots near the
> boundary itself.
> Hence if the intensity on one side of a boundary is near zero then the
> undershoots
> can create negative intensities after interpolation.
>
> You can intuitively see this by the fact that the sinc kernel itself
> goes negative,
> hence even interpolating a single spike (delta function) will introduce
> negative values.
> In contrast, trilinear always keeps the interpolated intensities between
> the maximum
> and minimum intensities, although it tends to pull them towards the mean
> intensity
> value, creating the known smoothing effect on the data.
>
> In practice, sinc normally doesn't introduce big negative values (if it
> is then there
> might be a bug) so you could just set them to zero using avwmaths and
> that would
> solve your undefined problems.
>
> Unfortunately, no interpolation method has all the desirable properties
> we would
> like! :(
>
> All the best,
> Mark
>
> Erik-Jan Vlieger wrote:
> >Hi,
> >
> >I used flirt to undo some of the damage induced by eddy currents which
> > occur during diffusion weighted imaging. What I did was a slice by slice
> > 2D affine match.
> >In order to keep the resolution as high as possible, I used the sinc
> >interpolation during the final reslice step. However, I noticed that the
> >resliced images contained negative values, whereas the input only
> > contained positive values. When I changed the interpolation step to
> > trilinear, these negative values did not occur.
> >Unfortunately, during FA and ADC reconstruction, negative values are
> >undefined. Therefore I cannot use the sinc-resliced images.
> >
> >We discussed the occurence of negative values, and concluded that it
> > should not be possible if the input only contains positive values. So we
> > think there might be a bug in the sinc interpolation module. Does anyone
> > agree?
> >
> >Thanks,
> >
> >Erik-Jan Vlieger
|