[vtk-developers] Problem writing out 'hollow' contours inrtstructs

[Dean Inglis]

Mark,

here are some possible solutions to 1 for now:

1) the ability to determine if one polydata is entirely contained
within another (I don't think that this is trivial, given how
contorted these polydata can be)

- a first rejection test: form bounding boxes (see vtkBoundingBox) and
do simple range inclusion tests:
    myvtkBoundingBox.AddBounds(myvtkPolyData.GetBounds())
    myvtkBoundingBox.Contains( anothervtkBoundingBox )
- triangulate the outer polygon and then do point inclusion tests
against the resulting triangles with the inner polygon vertices
- perform edge intersection tests

the last two options could be made somewhat efficient if you
make use of appropriate vtkLocator subclasses.

Dean

----- Original Message ----- 
From: "Mark Roden" <mmroden at gmail.com>
To: "Dean Inglis" <dean.inglis at camris.ca>
Cc: <vtk-developers at vtk.org>
Sent: Tuesday, December 21, 2010 11:29 AM
Subject: Re: [vtk-developers] Problem writing out 'hollow' contours 
inrtstructs


Hi Dean,

Yes, this is it exactly, and how Eclipse handles this case.  As to how
they do it, as far as I know, their source is not opened.

One possible solution is to identify those two polygons and look for
the 'leftmost' or 'topmost' points in each (ie, the point with the
lowest x value or the lowest y value), but that assumes that the two
polygons have been sampled identically-- if one were drawn such that
the underlying points were spaced 10 pixels apart, and the other 2,
then the 'leftmost' or 'topmost' might not line up.  If there were a
way to arbitrarily resample a polydata to ensure single-pixel spacing,
then that sampling issue would be solved.

Another possible problem is folds in the outer polygon.  If the outer
polygon has a 'ripple' near the leftmost/topmost point, then the
keyhole shouldn't be opened with the leftmost point in the polygon,
but with the closest point.

So, with that in mind, I'd need two functions:
1) the ability to determine if one polydata is entirely contained
within another (I don't think that this is trivial, given how
contorted these polydata can be)
2) Joining two polydata with a keyhole, using the 'topmost' point of
the inner polydata to the closest point on the outer polydata

Thanks,
Mark

On Mon, Dec 20, 2010 at 5:34 PM, Dean Inglis <dean.inglis at camris.ca> wrote:
> Hi Mark,
>
> I think if you have a donut for example, and want
> to encode it you would have to form the donut
> by bringing the free ends of a "C" arbitrarily close
> together so that you really only have one vtkPolyData
> object. I think if this example of a rectangle with a hole
> reproduces correctly in the list archive, you'll see what I
> mean:
>
> 0----------1
> | 5---4 |
> | | | |
> | | 3---2
> | | 8---9
> | | | |
> | 6---7 |
> 11---------10
>
>
> points 3, 8 and 2, 9 are coincident and form the keyhole.
> I this how Eclipse accomplishes this? The problem still
> remains how to cut arbitrary polydata automatically:
> maybe by finding the inner polygon centroid and
> emanating a radial outward?
>
> Dean
>
> ----- Original Message ----- From: "Mark Roden" <mmroden at gmail.com>
> To: <vtk-developers at vtk.org>
> Sent: Monday, December 20, 2010 7:25 PM
> Subject: [vtk-developers] Problem writing out 'hollow' contours in 
> rtstructs
>
>
> Hi all,
>
> I've got a problem writing out RTStruct objects using GDCM. The
> problem isn't with GDCM or VTK as they stand now, but rather with a
> particular section of the DICOM standard (3.3-2009 C 8.8.6.3). The
> problem has to do with a 'hollow' contour, or a polygon entirely
> encompassed within another polygon.
>
> To quote:
> When a single ROI describes an excluded inner volume, this can be
> encoded with a single contour, using a “keyhole” technique. In this
> method, an arbitrarily narrow channel is used to connect the outer
> contour to the inner contour, so that it is drawn as a single contour.
> An example of such a structure is shown in Figure C.8.8.6-1.
>
> If I have a set of vtkPolyData objects on a particular plane, is there
> an easy way to accomplish the above functionality?
>
> As I understand it, vtkPolyData are (or can be) oriented glyphs that
> indicate how a polygon should be drawn. If I were to just connect any
> two polydata objects, I'd imagine that would would need to be
> reversed. Is that understanding correct?
>
> Or if I made a binary representation of the mask that had a hollow in
> it, is there a way to indicate to the marching cubes to use the
> keyhole to join the inner and outer polygons?
>
> Part of the problem of this keyhole technique is the use of the words
> 'can be encoded'-- so while it's possible that other schemes could be
> used, this scheme is the only one suggested by the standard and is the
> one supported by Eclipse (and as far as I know, Eclipse's
> implementation is the one to match at the moment).
>
> Thanks,
> Mark
> _______________________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Follow this link to subscribe/unsubscribe:
> http://www.vtk.org/mailman/listinfo/vtk-developers
>
>
_______________________________________________
Powered by www.kitware.com

Visit other Kitware open-source projects at 
http://www.kitware.com/opensource/opensource.html

Follow this link to subscribe/unsubscribe:
http://www.vtk.org/mailman/listinfo/vtk-developers