[Insight-developers] A possible bug in neighborhood interator or filter?

[Paul Yushkevich]

Hi Luis,

I believe that the line
ImageRegion<3> region = image->GetLargestPossibleRegion();

initializes all the components of the region.  The subsequent calls to 
region.SetSize() and region.SetIndex() shrink the region in one of the 
dimensions to be one pixel thin.  When this dimension is in the X axis, 
everything works, but if it's in a different axis, then the code crashes.

What I'm trying to do is to perform only as much blurring as is 
necessary to produce one slice in the output image.

I think that there is some sort of a bug here.  The DiscreteGaussian 
filter chains three Neighborhood filters together, and I think that the 
regions produced by these filters might not be being calculated 
correctly.  That's my intuition after some debugging.

Thank you!

Paul

Luis Ibanez wrote:

> Hi Paul,
>
> The methods
>
>   SetSize(  component, value  )
>   SetIndex( component, value  )
>
> set only one component of the Size/Index.
>
> In the ImageRegion() default constructor
> both m_Size and m_Index are filled up with
> zeros.
>
> In your example, since you are initializing
> only the second component of the size, the
> first one is still zero. So, you seem to be
> asking for a region of size:
>
>      size =  (0,1)
>
>      0 along X
>      1 along Y
>
> Most iterators will be confused with this size.
> They expect to have a value of at least 1 along
> each dimension.
>
> It is actually surprising that it works with
>
>    region.SetSize( 0, 1 );
>
>
> You may want to call something like
>
>    region.SetSize(0, sizeX );
>    region.SetSize(1, sizeY );
>    region.SetSize(2, sizeZ );
>
>    region.SetIndex(0, indexX );
>    region.SetIndex(1, indexY );
>    region.SetIndex(2, indexZ );
>
>
> Since your image is 3D
>
>
>
>    Luis
>
>
>
> -----------------------
> Paul Yushkevich wrote:
>
>> The following code throws an exception in line 36 of 
>> itkZeroFluxNeumannBoundaryCondition.txx when filter->Update() is called.
>>
>> #include "itkImageFileReader.h"
>> #include "itkImage.h"
>> #include "itkDiscreteGaussianImageFilter.h"
>>
>> using namespace itk;
>>
>> void main(int argc, char *argv[])
>> {
>>
>>  // Typedefs
>>  typedef Image<float,3> ImageType;
>>  typedef ImageFileReader<ImageType> ReaderType;
>>  typedef DiscreteGaussianImageFilter<ImageType,ImageType> FilterType;
>>
>>  // Load image
>>  ReaderType::Pointer reader = ReaderType::New();
>>  reader->SetFileName("MRIcrop-orig.gipl");
>>  reader->Update();
>>  ImageType::Pointer image = reader->GetOutput();
>>
>>  // Create a requested region
>>  ImageRegion<3> region = image->GetLargestPossibleRegion();
>>  region.SetIndex(1,55);
>>  region.SetSize(1,1);
>>
>>  // Create filter
>>  FilterType::Pointer filter = FilterType::New();
>>  filter->SetVariance(1.0f);
>>  filter->SetInput(image);
>>  filter->GetOutput()->SetRequestedRegion(region);
>>  filter->Update();
>> }
>>
>> I didn't attach the image because it's a bit large for posting to the 
>> list.  However, it is of sufficient size (78x110x64). It seems that 
>> the neighborhood iterator jumps outside of bounds, causing the 
>> exception.  The funny thing is that it works with
>>
>>  region.SetIndex(0,55);
>>  region.SetSize(0,1);
>>
>> but not with
>>
>>  region.SetIndex(2,55);
>>  region.SetSize(2,1);
>>
>> Paul.
>>
>
>
>
>


-- 
--------------------------------
Paul A. Yushkevich, Ph.D.
President, Cognitica Corporation

17 Flemington Rd
Chapel Hill, NC 27517
Tel: 1-919-929-7652
--------------------------------