[vtk-developers] The ImageStencil Issue

[David Gobbi]

Karthik Krishnan wrote:
> Karthik Krishnan wrote:
>
>> Hi David,
>>
>> I find stencils very useful. I've been using them as a data structure 
>> for representing segmentations.
>>
>> One of the things I would like to contribute to stencils is minkowski 
>> operators or "set" operations on stencils. For instance union of two 
>> stencils, stencilA-stencilB, intersection of stencils etc.. The need 
>> for these arises when a 3D volume has a few disconnected hand-drawn 
>> segmentations and as a user, I would like to discard a segment etc.. 
>> I've been using these separately in a derived class, but I could move 
>> them to the vtkImageStencilData class itself.
>
> They've have been added to the vtkImageStencilData class.
>
> Thanks
> -karthik

Great, that should do it.  You don't also happen to have 
vtkAlgorithm-type class for pipelining these operations, do you?  If 
not, I can add it to the wish list for stencils.

 - David


>> Another addition, I would hope to see would be grayscale stencils for 
>> fuzzy segmentations. Imagine a brush widget dabbing a fuzzy / 
>> dithered stroke and you would like to use the stencildata to 
>> represent opacity. It would be great if the stencils could store a 
>> DataArray in addition to maintaining the ExtentLists. This could 
>> possibly be done in a sublcass.
>>
>> Thanks
>> -karthik
>>
>>
>> David Gobbi wrote:
>>
>>> Hi All,
>>>
>>> I'd like to redress some ugliness that I introduced to VTK a few 
>>> years back.
>>>
>>> The vtkImageStencilSource class requires the Spacing and Origin of 
>>> whichever image eventually uses the stencil to be propagated the 
>>> wrong way along the pipeline.  Ouch!
>>>
>>> Obviously the vtkImageStencilSource should actually take an Image as 
>>> an input and should then use that image to get the Spacing and 
>>> Origin to use for the stencil.  This is a fairly minor change.  For 
>>> VTK 5.2, I can then make VTK produce a deprecation warning if people 
>>> don't set an input.
>>>
>>> I would call the method SetInformationInput() to make it clear that 
>>> only the information (the Spacing, Origin, and WholeExtent) of the 
>>> data would be used.  Also, vtkPolyDataToImageStencil() needs it 
>>> SetInput() method for the polydata input.
>>>
>>> For example:
>>>
>>> stencilSource = vtkImplicitFunctionToImageStencil()
>>> stencilSource.SetInput(vtkSphere())
>>> stencilSource.SetInformationInput(input)
>>>
>>> histogram = vtkImageAccumulate()
>>> histogram.SetStencil(stencilSource.GetOutput())
>>> histogram.SetInput(input)
>>>
>>> Do people think this is a good change?  Are there many people out 
>>> there who even use stencils in VTK?  How many people know what a 
>>> stencil is?  (Hint: it is an image mask stored in an efficient 
>>> run-length format instead of vtkImageData format).
>>>
>>> Also, I'm not sure what the methods should be named in light of the 
>>> VTK 5 pipeline changes.   Maybe I should name them as follows:
>>>
>>> SetInformationInputConnection(vtkAlgorithmInput *) instead of 
>>> SetInformationInput(vtkImageData *)
>>> SetStencilInputConnection(vtkAlgorithmInput *) instead of 
>>> SetStencil(vtkImageStencilData *)
>>>
>>> I'd like to get this right, and I think that only small changes are 
>>> necessary to do so.
>>>
>>> Also, of course, I'd like to see more of the filters in Imaging 
>>> utilizing stencils.  Right now only a handful do (Blend, Reslice, 
>>> Stencil, Accumulate).  The potential here is very high... a stencil 
>>> allows you to perform an operation only within a specified region of 
>>> interest, where the shape of the region can be whatever you desire.  
>>> The ROI can even be described by a vtkPolyData mesh or by a 
>>> vtkImplicitFunction, since these types can be converted efficiently 
>>> into vtkImageStencilData.
>>>
>>> - David
>>>