[Insight-developers] GaussianDerivativeImageFunction does not
transform points from physical space to pixel space in Evaluate(..)
Dan Mueller
d.mueller at qut.edu.au
Wed Feb 15 23:25:44 EST 2006
Hi all,
For my application I need to calculate the image derivative at a number
of physical locations. I have found two suitable ImageFunctions:
CentralDifferenceImageFunction and GaussianDerivativeImageFunction. The
GaussianDerivativeImageFunction is more suitable to my application
because it uses a Gaussian kernel to smooth noise artifacts.
Unfortunately, there seems to be a discrepancy with the
GaussianDerivativeImageFunction::Evaluate(const PointType& point)
function. The Evaluate(const PointType& point) function should evaluate
the function at the given physical point (ie. taking into account the
image spacing and origin). For example see
itkCentralDifferenceImageFunction.h which - via
itk::ImageFunction::ConvertPointToNearestIndex(..) - uses
itk::Image::TransformPhysicalPointToContinuousIndex(..) to transform the
physical point to a ContinuousIndex in pixel space before evaluation.
It appears that GaussianDerivativeImageFunction *does not* transform the
point from physical space to pixel space before evaluation. If you look
at the Evaluate(..) function in the
itkGaussianDerivativeImageFunction.txx file, the physical point is cast
to an index (the rounding differences are used as the offset to
recompute the kernel) and then itk::Image::GetPixel(const IndexType
&index) is invoked - no transformation from physical space to pixel
space occurred!
Have I missed something, or is this an omission?
Thanks for your help.
Dan
--
Dan Mueller (d.mueller[at]qut.edu.au)
School of Engineering Systems (ES)
Faculty of Built Environment and Engineering (BEE)
Queensland University of Technology (QUT)
CRICOS No: 00213J
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.itk.org/mailman/private/insight-developers/attachments/20060216/b9ed7190/attachment.html
More information about the Insight-developers
mailing list