[vtkusers] Voxelize a point cloud (vtkVoxelModeller?)

[lynx.abraxas at freenet.de]

On 02/11/09 17:18:41, David Doria wrote:
> Lynx,
>
> Yes, this is perfect. I made an example here:
> http://www.vtk.org/Wiki/Embed_points_into_a_volume
> Maybe vtkVoxelModeller should be marked as deprecated? Or was I not
> using it for the its intended use?
>
> I tried to do SetRadius(0) which I thought would set the "spread" of
> the point to be zero (an impulse) and therefore the entire
> contribution of the point would be added directly to the voxel that it
> is inside. However, when I did that (and even with SetRadius(.01)),
> none of the voxels received any contributions (i.e. the resulting
> imagedata was completely zeros). Do you know why that might be?
>

Hm,  I'm  not  sure.  I've  been  wondering about the parameters the Gaussiain
Splatter needs.
(there is an example http://www.itk.org/Wiki/Create_a_surface_from_Unorganized_Points_(Gaussian_Splat))
The  radius (as the docs state, SetRadius) is expressed as a percentage of the
length of the longest side of the sampling volume. I haven't found the time to
check  but  perhaps 1/(2n) (eg. 0.05) would give an effective range of 1 voxel
in Your example. Below (if rounded) will probably result in what You got.
I'm not really  understanding  SetModelBounds.  I'd  think  a  parameter  that
restricts  the  sampling  reagion  is  necessary since the Gaussian extends to
infinity. If SetModelBounds
is doint that setting that to 1 voxel might help as well.
One thing I'm not sure is if the filter generates a Gaussian  blob  (extending
acc.  to  ModelBounds)  at  the  injection  point  or  if  it "injectes random
secondary points" distributed gaussian. In the latter (random) case one  would
need to worry if there is always a full contribution to the centre voxel.

Hope this might be of help.
Lynx