[Insight-developers] GradientVectorFlowImageFilter internal image type

[Karthik Krishnan]

Gaëtan Lehmann wrote:
>
> Le 10 janv. 09 à 16:20, Karthik Krishnan a écrit :
>
>> Gaëtan Lehmann wrote:
>>>  template <class TInputImage, class TOutputImage, class 
>>> TInternalPixel=double>
>>>  class ITK_EXPORT GradientVectorFlowImageFilter : public 
>>> ImageToImageFilter<TInputImage, TOutputImage>
>>>
>>> There shouldn't be any backward compatibility issue that way. I 
>>> don't know well that class though, so I have an important question:
>>> Does that change look reasonable? I mean, is it relevant to use a 
>>> different internal pixel type than double - float for example?
>> I don't think it makes sense to use the class with any internal pixel 
>> type other than float or double. Actually, the input images (and 
>> output images) supplied should also be float/double. The filter 
>> internally computes the laplacian image using the internal pixel 
>> type. Double precision is also needed to account for the diffusion of 
>> the gradient vectors.
>>
>> I would just wrap the class for double and float alone.
>
> My problem comes from this method:
>
>     itkSetMacro(LaplacianFilter, LaplacianFilterPointer);
>
> The LaplacianFilter currently uses double as pixel type, whatever are 
> the input or the output image types, so that method expose a filter 
> type which is not available most of the time, because the default real 
> type wrapped in wrapitk is float. The result is a warning at build 
> time and an unusable SetLaplacianFilter() in the target languages. 
> Changing the internal pixel type (used in the laplacian filter type) 
> to the one used in the input and the output image would avoid the 
> problem with the SetLaplacianFilter() method.
> An easy way to fix the problem, but if it produce wrong result because 
> the double precision is required for the internal image type, that's 
> also a wrong way to fix that problem.
>
> With that fix, I would wrap the types
>
>   itk::GradientVectorFlowImageFilter< itk::Image<float, 2>, 
> itk::Image<float, 2>, float >
>   itk::GradientVectorFlowImageFilter< itk::Image<float, 3>, 
> itk::Image<float, 3>, float >
>
> by default, and/or
>
>   itk::GradientVectorFlowImageFilter< itk::Image<double, 2>, 
> itk::Image<double, 2>, double >
>   itk::GradientVectorFlowImageFilter< itk::Image<double, 3>, 
> itk::Image<double, 3>, double >
>
> if the user has chosen to wrap double.
> But does it make sense to have itk::Image<float, 2> used for the 
> laplacian, when input and output image are itk::Image<float, 2>?
Sure. It makes sense to wrap the following 4 combinations alone.

  itk::GradientVectorFlowImageFilter< itk::Image<float, 2>, 
itk::Image<float, 2>, float >
  itk::GradientVectorFlowImageFilter< itk::Image<float, 3>, 
itk::Image<float, 3>, float >
  itk::GradientVectorFlowImageFilter< itk::Image<double, 2>, 
itk::Image<double, 2>, double >
  itk::GradientVectorFlowImageFilter< itk::Image<double, 3>, 
itk::Image<double, 3>, double >


Thanks
--
karthik

>
>>
>>
>> thx
>> -- 
>> karthik
>>>
>>> Regards,
>>>
>>> Gaëtan
>>>
>>>
>>> ------------------------------------------------------------------------ 
>>>
>>>
>>> _______________________________________________
>>> Insight-developers mailing list
>>> Insight-developers at itk.org
>>> http://www.itk.org/mailman/listinfo/insight-developers
>>>
>>
>> -- 
>> Karthik Krishnan
>> R & D Engineer,
>> Kitware Inc,
>> Ph:  518 371 3971 x119
>> Fax: 518 371 3971
>>
>

-- 
Karthik Krishnan
R & D Engineer,
Kitware Inc,
Ph:  518 371 3971 x119
Fax: 518 371 3971