[Insight-users] runtime warnings when hooking up hessian filter toeigenanalysis2d filter.

[Andreas Keil]

Dear Jorik (and others),

did you already solve your problem? I would also like to construct a 2D
vessel filter by combining HessianRecursiveGaussian and EigenAnalysis2D...

Any help would be really appreciated since I cannot find a better way to
to curvature-based image analysis in 2D.

Thank you,
Andreas.


-----Original Message-----

Hi Max,

ImageAdaptors are somehow incompatible with dynamic_cast test.

In general, using dynamic_casts is a violation of the basic principles of
 Generic Programming. However, we have been forced to use it in several
places, for example, when managing itkEventObject.

In the case that you are confronting, the use of ImageAdaptors happens
to be incompatible with the dynamic_casts that the ITK code is attempting.

In order to get rid of the warnings, you may have to use a real ITK Filter
instead of an ImageAdaptor. In that way, the input image to the
EigenAnalysis2DImageFilter will actually correspond to the type that
your are setting in its template parameters.

Another option is for you to modify the code of the
EigenAnalysis2DImageFilte
to accept a Hessian image, instead of expecting the three components of
the
Hessian to be passed as independent images. For an example on how to do
this, you may want to look at the code of the

Hessian3DToVesselnessMeasureImageFilter

http://public.kitware.com/Insight/Doxygen/html/classitk_1_1Hessian3DToVess
elnessMeasureImageFilter.html

      Regards,
           Luis



-----Original Message-----
Sent: Thursday, March 15, 2007 11:52
To: insight-users at itk.org
Subject: [Insight-users] runtime warnings when hooking up hessian filter
toeigenanalysis2d filter.

Hello,

In my image processing pipeline, I have to generate the eigenvalues of the
hessian.  I'm not a long-time ITK user, but I figured the combination of
Itk::HessianRecursiveGaussianFilter and an itk::EigenAnalysis2DImageFilter
were the best components to use.  I've put an nth-component adapter
inbetween to select the right parts of the hessian matrix to feed to the
three inputs of the eigen analysis filter. 

Everything compiles and runs well now, except for a few pesky warnings at
runtime:

WARNING: In
/usr/local/include/InsightToolkit/BasicFilters/itkEigenAnalysis2DImageFilt
er.txx, line 133
ImageToImageFilter (0x689390):
EigenAnalysis2DImageFilter::GetMaxEigenVector(): dynamic_cast has failed.
A reinterpret_cast is being attempted. 
Type name is: N3itk5ImageINS_6VectorIfLj2EEELj2EEE

Running this trough demangle with a _Z prepended showed that this was of
type
itk::Image<itk::Vector<float, 2u>, 2u>

I feel this has something to do with the floatvectorimage that the
eigenvalue filter is outputting perhaps, but I cannot find the cause of
this warning. 


My current processing pipeline is as follows:


typedef float FloatPixelType; 

typedef itk::Image<FloatPixelType, Dimension> FloatImageType;

typedef itk::HessianRecursiveGaussianImageFilter< 
FloatImageType> HessianFilter;

typedef itk::Image<
itk::Vector< FloatPixelType, 2 >, 
Dimension > FloatVectorImageType;

typedef itk::EigenAnalysis2DImageFilter< 
FloatImageType, 
FloatImageType, 
FloatVectorImageType> EigenAnalysis2DImageFilterType;

// to make the output of the hessian filter readable by the
// eigenanalysis filter, we need to adapt the image 
typedef itk::NthElementPixelAccessor< 
FloatPixelType,
HessianFilter::OutputPixelType > OutputImageAdaptorType;

typedef itk::AdaptImageFilter< 
HessianFilter::OutputImageType, 
FloatImageType, 
OutputImageAdaptorType > AdaptType;


// the hessian filter is hooked up to the original input filter
HessianFilter::Pointer hesFilter = HessianFilter::New();
hesFilter->SetInput( readCast->GetOutput() ); 

EigenAnalysis2DImageFilterType::Pointer eigFilter =
EigenAnalysis2DImageFilterType::New();

// make the three components of the hessian to the eigFilter
AdaptType::Pointer adapterE0 = AdaptType::New(); 
adapterE0->SetInput( hesFilter->GetOutput() );
adapterE0->GetAccessor().SetElementNumber( 0 );

AdaptType::Pointer adapterE1 = AdaptType::New();
adapterE1->SetInput( hesFilter->GetOutput() ); 
adapterE1->GetAccessor().SetElementNumber( 1 );

AdaptType::Pointer adapterE2 = AdaptType::New();
adapterE2->SetInput( hesFilter->GetOutput() );
adapterE2->GetAccessor().SetElementNumber( 2 ); 

// and hook up the adapted images to the eigen analysis filter.
eigFilter->SetInput1( adapterE0->GetOutput() );
eigFilter->SetInput2( adapterE1->GetOutput() );
eigFilter->SetInput3( adapterE2->GetOutput() ); 


And then the eigFilter's getOutput(0) and getOutput(1) are used in further
processing.

I hope someone can help to shed some light on these warnings.

Thanks,
Jorik