[MINC-users] mincresample and slice-based scaling

[Andrew Janke]

Hi Alex,

My guess is that this is a combination of discretisation (of your
chosen data type) and the macros hidden deep within MINC that do the
tri-linear interpolation. Probably more of the former than the latter
though.

Does the same problem happen when you use a float/double volume?

In any case the discretisation is very minimal and I'd say within
expected values given the range of a short (5 places of precision)
matches the error you are seeing.


a

On 21 March 2013 03:43, Alex Zijdenbos <zijdenbos at gmail.com> wrote:
> Hi all,
>
> I came across something somewhat unexpected (at least to me), and was
> wondering if anybody would be able to clarify this:
>
> I have a volume with a real range [0,255] (signed short, the valid
> range=full range of data type, no slice scaling). When I mincresample that
> using a linear xfm and some new starts and steps, I end up with a global
> range (from mincstats) of [-0.0002770616653,255.0002071]. I am used to
> seeing over- and undershoots with tricubic interpolation, but I was always
> under the impression that trilinear by definition should maintain the
> extrema. When I add the -keep_real_range flag to the mincresample call, the
> resulting range is the expected [0,255] (again from mincstats); so this
> essentially fixes the problem. But why would the voxel range change when
> using trilinear interpolation? Precision issues? Slice scaling issues?
> Extrapolation instead of interpolation perhaps?
>
> -- A
> _______________________________________________
> MINC-users at bic.mni.mcgill.ca
> http://www.bic.mni.mcgill.ca/mailman/listinfo/minc-users