[Insight-developers] Image Derivatives - physical vs. index space

[sgerber at cs.utah.edu]

What I meant is that the output of the itkDerivativeImageFilter is a  
scalar image and not a vector image and it seems that calling
TransformLocalVectorToPhysicalVector
would require to package the scalar into a vector which would be a lot  
of unnecessary overhead.

Regards
Sam

Quoting Luis Ibanez <luis.ibanez at kitware.com>:

> On Thu, Jul 2, 2009 at 6:01 PM, Samuel Gerber <sgerber at cs.utah.edu> wrote:
>
>> Samuel Gerber wrote:
>>
>>> Hi Luis
>>> <http://public.kitware.com/Bug/my_view_page.php>
>>>
>>>> The fix will come down to calling the image method:
>>>>
>>>>  inputImage->RotateArrayByDirectionCosines( gradient, rotatedGradient );
>>>>
>>>> after every computation of a derivative vector.
>>>>
>>> I would assume, since the derivative is axis aligned in index space, it
>>> could be simplified by only multiplying with the row corresponding to the
>>> SetDirection setting in the itkDerivativeImageFilter.
>>>
>> -------------------------
>
>
> I believe that this is what the RotatArrayByDirectionCosines() method
> does internally, but of course, there is noting better than looking at
> the source code   :-)
>
>
> BTW. This method was later renamed as
>
>     TransformLocalVectorToPhysicalVector()
>
> You will find it in
>
>      Insight/Code/Common/itkImageBase.txx
>
> in lines 498-524:
>
>
>   template<class TCoordRep>
>   void TransformLocalVectorToPhysicalVector(
>     const FixedArray<TCoordRep, VImageDimension> & inputGradient,
>           FixedArray<TCoordRep, VImageDimension> & outputGradient ) const
>     {
>     //
>     // This temporary implementation should be replaced with Template
> MetaProgramming.
>     //
> #ifdef ITK_USE_ORIENTED_IMAGE_DIRECTION
>     const DirectionType & direction = this->GetDirection();
>     for (unsigned int i = 0; i < VImageDimension; i++)
>       {
>       typedef typename NumericTraits<TCoordRep>::AccumulateType
> CoordSumType;
>       CoordSumType sum = NumericTraits<CoordSumType>::Zero;
>       for (unsigned int j = 0; j < VImageDimension; j++)
>         {
>         sum += direction[i][j] * inputGradient[j];
>         }
>       outputGradient[i] = static_cast<TCoordRep>( sum );
>       }
> #else
>     for (unsigned int i = 0; i < VImageDimension; i++)
>       {
>       outputGradient[i] = inputGradient[i];
>       }
> #endif
>     }
>
>
>
>
> I would discourage exposing the computation, since then we will end
> up with repeated implementations of the code in multiple locations.
>
>
> In other words, you shouldn't need to worry about how the gradient
> vector is rotated. (unless we suspect of a bug in that code....)
>
>
>
>       Regards,
>
>
>                 Luis
>
>
> -------------------------------
>
>>
>>>
>>>  In the meantime,
>>>> Could you use the itkGradientImageFilter, instead
>>>> of the DerivativesImageFilter for your application ?
>>>>
>>> Yes I can use the gradient filter instead.
>>>
>>>
>>> Thanks
>>> Sam
>>>
>>>
>> _______________________________________________
>> Powered by www.kitware.com
>>
>> Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html
>>
>> Please keep messages on-topic and check the ITK FAQ at:
>> http://www.itk.org/Wiki/ITK_FAQ
>>
>> Follow this link to subscribe/unsubscribe:
>> http://www.itk.org/mailman/listinfo/insight-developers
>>
>



----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.