[Insight-users] gradient calculation

[Matt McCormick]

Hi Cagatay,

> I am able to obtain better results for my
> problem on synthetic data with equal spacing.

I may have misinterpreted your initial email.  ITK handles anisotropic
spacing very well.  You certainly do not need equal spacing in every
direction.  The point I was trying to make was the spacing between
points in a given direction must be constant, i.e. an 'ImageGrid' as
opposed to a 'RectilinearGrid' as described here:

http://github.enthought.com/mayavi/mayavi/data.html


> This filter has a function SetUseImageSpacing. What I am wondering
> is the relationship between the spacing and the sigma value I
> should be providing. The other parameters are making perfect
> sense to me, as I decrease the maximum error parameter the
> kernel width should going to increase.
> As an example:
> If for my synthetic data with equal spacing (1,1,1), I use sigma=3,
> should I be using sigma=1.5 for a spacing of  (0.5,0.5,0.5) with
> UseImageSpacing turned on?  It looks like the filter internally
> divides the spacing with the spacing when this flag is on.
> I would appreciate insights to this function.

Yes, if UseImageSpacing is On, you can specify sigma in physical
units.  That is, if the content of the image is not changing, just how
it is sampled, the sigma will be the *same* to have the same effect.
Even if the spacing is not isotropic, the same physical sigma in each
direction can be specified to have the same physical effect in every
direction.

Hope this helps,
Matt