[vtkusers] BUG: vtkImageData::ComputeBounds() ... + solution

[Ken Martin]

> maybe I have a different understanding of the concept
> "bounds", but I think the current implementation of
> vtkImageData::ComputeBounds() isn't quite right. (vtk version 4.0)
>
> void vtkImageData::ComputeBounds()
> {
>   float *origin = this->GetOrigin();
>   float *spacing = this->GetSpacing();
>
>   this->Bounds[0] = origin[0] + (this->Extent[0] * spacing[0]);
>   this->Bounds[2] = origin[1] + (this->Extent[2] * spacing[1]);
>   this->Bounds[4] = origin[2] + (this->Extent[4] * spacing[2]);
>
>   this->Bounds[1] = origin[0] + (this->Extent[1] * spacing[0]);
>   this->Bounds[3] = origin[1] + (this->Extent[3] * spacing[1]);
>   this->Bounds[5] = origin[2] + (this->Extent[5] * spacing[2]);
> }
>
>
> I think it should be replaced by:
>
> void vtkImageData::ComputeBounds()
> {
>   float *origin = this->GetOrigin();
>   float *spacing = this->GetSpacing();
>
>   this->Bounds[0] = origin[0] + (this->Extent[0] * spacing[0]);
>   this->Bounds[2] = origin[1] + (this->Extent[2] * spacing[1]);
>   this->Bounds[4] = origin[2] + (this->Extent[4] * spacing[2]);
>
>   this->Bounds[1] = origin[0] + ((this->Extent[1] + 1) *
> spacing[0]);
>   this->Bounds[3] = origin[1] + ((this->Extent[3] + 1) *
> spacing[1]);
>   this->Bounds[5] = origin[2] + ((this->Extent[5] + 1) *
> spacing[2]);
> }
>
>
> Just to clarify this with an example...
> suppose I would create a piece of vtkImageData with
> dimensions 32x32x32.
> This would give me an extent of [0, 31, 0, 31, 0, 31]
> I set the spacing to (10, 10, 10) and the origin to (0, 0, 0)
>
> IMHO this would give me bounds of [0, 320, 0, 320, 0, 320],
> but the current implementation yields [0, 310, 0, 310, 0, 310]
>
> If you agree, could this be corrected?

This is a difficult issue and one I have raised in the past. Currently
vtk defines an image to only exist between sample points. As such its
bounds are correct. This has a number of advantages for mixing images
and polygonal meshes. Few people would suggest that the bounds of a
polygonal mesh go outside of the sample points. But, another
interpretation of an image is to represent the area around a sample as
a pixel. in which case the image bounds start a half pixel before the
first sample and end a half pixel after the last sample. And the third
interpretation is what you suggested, that the sample represents the
lower left corner of a pixel, so the bounds go from the first sample
to one pixel beyond the last sample. My suggestion would be the second
option; the samples represent essentially cell data, not point data.
But switching the vtkImage to use cell data would be a huge change.
The other option would be to modify the readers to read in cell data
and convert it to point data (so that a 256 by 256 pixel image would
result in a 257 by 257 point image) But that's a little confusing as
well. I'm open to suggestions, but so matter what the fix it goes
beyond just changing the bounds computation. At a minimum the
rendering must be changed as well so that the image is rendered into
the "correct" space.

Thanks
Ken