Neighborhood / cutoff distance of smoothing with itkSmoothingRecursiveGaussianImageFilter


(Philip Cook) #1

In the itkGaussianInterpolateImageFunction filter, we set sigma, the standard deviation of the Gaussian smoothing and alpha, the cutoff distance from the center voxel, over which the kernel is sampled.

Is there an equivalent parameter to alpha for itkSmoothingRecursiveGaussianImageFilter?


(Bradley Lowekamp) #2

The implementations of the Gaussian smoothing is done quite differently. The Gaussian interpolator uses a N-D weighted region to perform the convolution. This is similar to the DiscreteGaussianImageFilter, which has a similar a MaximumError and MaximumKernelWidth. The convolution will be done with N separated 1-D kernels.

The recursive Gaussian filters use a… recursive algorithm which does a 1-D forward processed to imaginary space, and a backwards processing step. The result is closer to “infinite” support of an approximation to a Gaussian. The doxygen has some links to some papers, but you may find other interesting reads about recursive Gaussian implementations other places.

Here is another recursive implementation available as a remote module: http://www.insight-journal.org/browse/publication/896

To answer your questions: no.


(Philip Cook) #3

Thanks @blowekamp.