Dear Stian,
> Thanks for the reply.
> The bicubic interpolation of Matlab's imtransform is AFAIK a
> Lagrangian interpolation that is used with the backward transform,
> but I guess cubic splines would already be a huge improvement over
> the bilinear interpolation that FLIRT offers.
I would actually be surprised of the Lagrange interpolation performed
better than the cubic B-spline, but I haven't compared them so I
cannot say for certain.
There are freely available code for spline interpolation both in FSL
(up to 7th order if I recall correctly) and in SPM. It would be easy
enough for you to incorporate either in your own code (FSL easiest if
you are using C++, SPM if you use Matlab) and then you would have
identical interpolation for all your methods. And both those are
"proper" 3D implementations.
> I tried the command you gave me, but applywarp complains because I
> didn't specify a "warp/coefficient volume". Also, "spline" isn't
> among the interpolation options (only: "{nn,trilinear,sinc}").
>
> Even though it seems to be a very recent version (4.1.4), I think I
> should add that I'm using the FSL package from the Ubuntu 9.10
> repositories.
True, you will need the latest version to avoid having to supply a
zero warp-field.
Good Luck Jesper
|