[Insight-users] 1D Convolution with Gaussian Kernel using

Antonio Gómez Barquero agb1 at alu.upct.es
Fri Nov 18 07:17:47 EST 2011 


Hi Nicholas,

Thanks so much for your answer, is helping me quite a lot.

Still have some difficulties... I use the NeighborhoodOperatorImageFilter,
it compiles quite fast, but the output is wrong. I only have results on the
first row, and they have just the opposite behavior that the one I was
expecting ( i mean, I expected then numbers go decreasing to zero with
x-increasing but they just increase until reaching the value '1' .

Just one question about the 'setVariance' that is not documented in the help
of the class... is it in terms of 'mm' or pixels? If it's not in terms of
pixels...what should I do for having it in that way?

Thanks so much again!!!

Antonio

-----Mensaje original-----
De: Nicholas Tustison [mailto:ntustison at gmail.com] 
Enviado el: jueves, 17 de noviembre de 2011 18:50
Para: Antonio Gómez Barquero
CC: 'Cory Quammen'; insight-users at itk.org
Asunto: Re: [Insight-users] 1D Convolution with Gaussian Kernel using

Hi Antonio,

Instead of using the ConvolutionImageFilter, you might want to explore
existing functionality in combining the GaussianOperator with the
NeighborhoodOperatorImageFilter.

  typedef itk::NeighborhoodOperatorImageFilter<ImageType, ImageType>
SmootherType;
  typename SmootherType::Pointer smoother = SmootherType::New();

  typedef GaussianOperator<PixelType, ImageDimension> GaussianType;
  GaussianType gaussian;
  gaussian.SetVariance( variance );
  gaussian.SetDirection( 0 );

  gaussian.SetMaximumError( 0.001 );
  gaussian.SetMaximumKernelWidth(
inputImage->GetRequestedRegion().GetSize()[0] );
  gaussian.CreateDirectional();

  smoother->SetOperator( gaussian );
  smoother->SetInput( inputImage );
  smoother->Update();

  smoothImage = smoother->GetOutput();

However, in case you want to keep your image, you can use
GaussianImageSource to create  2-D image which has size Nx1 and then that
image should plug right into the ConvolutionImageFilter.  I just did it
right now with my code and it seemed to work as expected.

Nick



On Nov 17, 2011, at 4:42 AM, Antonio Gómez Barquero wrote:

> 
> 
> Hi!
> 
> You were right, now it compiles, but I only have some numbers in the 
> first
> 20 positions, after that everything is 0, so is wrong, also it takes 
> around
> 15 seconds to perform the calculation which is a crazy thing!.
> 
> I prefer to generate the Gaussian myself and after that translate it 
> into an image ( kernel in this case) because I am doing work also with 
> Matlab, and I would like to use the same Gaussian.
> 
> But I think I am missing something in the body of the iteration , 
> something that performs the convolution itself or maybe the point of 
> doing the convolution in x-axis in 1D....
> 
>   while(!imageIterator.IsAtEnd()) 
>    {
>    //imageIterator.Set??
> 
>    ++imageIterator; //operator++  increments the iterator one position 
> in the positive direction
> 
>    }
> 
> 
> Also, the perfect final would be to have a Kernel of 25 pixels ( the 
> same I have originally in my Matlab program), and then performing the 
> convolution in each row of the x-direction of the image...I think that 
> must be a way...so I follow reading the ITK manual and let's see I found
something.
> 
> Any Tip or idea or help will be highly appreciated!
> 
> Antonio
> 
> -----Mensaje original-----
> De: Cory Quammen [mailto:cquammen at cs.unc.edu] Enviado el: miércoles, 
> 16 de noviembre de 2011 17:28
> Para: Antonio Gómez Barquero
> CC: insight-users at itk.org
> Asunto: Re: [Insight-users] 1D Convolution with Gaussian Kernel using
> 
> Antonio,
> 
> It looks like you are defining your kernel to be 256x256, not 256x1.
> You probably want to call
> 
> size[0] = width;
> size[1] = 1;
> 
> instead of
> 
> size.Fill(width);
> 
> You might want to consider using the GaussianImageSource
> (http://www.itk.org/Doxygen/html/classitk_1_1GaussianImageSource.html)
> to generate your Gaussian.
> 
> Hope that helps,
> Cory
> 
> 2011/11/16 Antonio Gómez Barquero <agb1 at alu.upct.es>:
>> SORRY, the last e-mail was icomplete, here is ok:
>> 
>> 
>> 
>> My goal is to calculate the calculation in each row of a 2D-image ( 
>> in the
>> x-direction)
>> 
>> After following the tip from Cory I am trying to use the 
>> ‘ConvolutionImageFilter’, and make a kernel with the Gaussian values 
>> I calculate before.
>> 
>> I took a look to this example
>> (http://www.itk.org/Wiki/ITK/Examples/ImageProcessing/ConvolutionImag
>> e
>> Filter) and my question is how to include the values of the Gaussian 
>> filter to the kernel, I tried wit the code below and it compiles, but 
>> it seems the execution has no end when I update the 
>> convolutionFilter
 Because of that I think that I am doing something 
>> wrong. My code is below and is not so much, if some can give some 
>> help I would appreciate that really a lot !!!!.
>> 
>> 
>> 
>>       typedef itk::ConvolutionImageFilter <ImageType> ConvFilterType;
>> 
>> 
>> 
>>       ImageType::Pointer kernel = ImageType::New();
>> 
>>       CreateKernel(kernel, Gaussian , 256);
>> 
>> 
>> 
>>       //Convolve image with kernel
>> 
>>       ConvFilterType::Pointer convolutionFilter = 
>> ConvFilterType::New();
>> 
>>       convolutionFilter->SetInput(Li_itk);
>> 
>>       convolutionFilter->SetImageKernelInput(kernel);
>> 
>>       convolutionFilter->Update();
>> 
>> 
>> 
>> void Reg_image_v2::CreateKernel(Reg_image_v2::ImageType::Pointer
>> kernel, double *Gaussian, unsigned int width) {
>> 
>> 
>> 
>>       ImageType::IndexType start;
>> 
>>       start.Fill(0);
>> 
>>       ImageType::SizeType size;
>> 
>>       size.Fill(width);
>> 
>>       ImageType::RegionType region;
>> 
>>       region.SetSize(size);
>> 
>>       region.SetIndex(start);
>> 
>>       kernel->SetRegions(region);
>> 
>>       kernel->Allocate();
>> 
>> 
>> 
>>       ImageType::IndexType pixelIndex;
>> 
>>       for (int i = 0 ; i < width ; i ++){
>> 
>>             pixelIndex[0] = i ;
>> 
>>             pixelIndex[1] = 0 ;
>> 
>>             kernel->SetPixel(pixelIndex,Gaussian[i]);
>> 
>>       }
>> 
>>       itk::ImageRegionIterator<ImageType> imageIterator(kernel, 
>> region);
>> 
>>    while(!imageIterator.IsAtEnd())
>> 
>>     {
>> 
>>     //imageIterator.Set??
>> 
>>     ++imageIterator; //operator++  increments the iterator one 
>> position in the positive direction
>> 
>>     }
>> 
>> 
>> 
>> 
>> 
>> } // end createKernel
>> 
>> 
>> 
>> 
>> 
>> Antonio Gómez Barquero
>> 
>> 
>> 
>> Ingeniero de Telecomunicaciones -Becario Investigador asociado a 
>> Actividades de I+D+I
>> 
>> GTTS ( Grupo de Tratamiento y Teoría de la Señal)[ 
>> http://gtts.upct.es/]
>> 
>> UPCT (Universidad Politécnica de Cartagena)[ http://www.upct.es/]
>> 
>> 
>> 
>> _____________________________________
>> Powered by www.kitware.com
>> 
>> Visit other Kitware open-source projects at 
>> http://www.kitware.com/opensource/opensource.html
>> 
>> Kitware offers ITK Training Courses, for more information visit:
>> http://www.kitware.com/products/protraining.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-users
>> 
>> 
> 
> 
> 
> --
> Cory Quammen
> Research Associate
> Department of Computer Science
> The University of North Carolina at Chapel Hill
> 
> _____________________________________
> Powered by www.kitware.com
> 
> Visit other Kitware open-source projects at 
> http://www.kitware.com/opensource/opensource.html
> 
> Kitware offers ITK Training Courses, for more information visit:
> http://www.kitware.com/products/protraining.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-users

More information about the Insight-users mailing list