[Insight-users] Fast convolution implementation?

[svoboda at fi.muni.cz]

Hi,

I'm a newbie in using ITK and I'd like to ask for help. I'd like to 
implement fast (efficient) convolution of two large 3D images. I thought I 
could use "itkConvolutionImageFilter" class but I found out that it does 
not use FFT, i.e. it performs the naive convolution algorithm using 
neighbourhood, which is quite slow. Am I right?

If so, I should probably use convolution theorem, i.e. apply FFT to the 
input image and also to the kernel, them multiply the results and 
subsequently apply inverse FFT. Is such a code (written in ITK) available 
somewhere? Thanks a lot!

With regards
David