FFTW vs OpenCV cvDFT

Question

Can acceleration be expected when using FFTW (http://www.fftw.org/) instead of OpenCV cvDFT (http://goo.gl/YCHj0)?

My program execution time is heavily determined by the reverse and direct DFT application, and I am thinking about using FFTW instead of OpenCV cvDFT.

IIRC FFTW does some kind of "compilation at runtime," while cvDFT is a simple FFT implementation, so I think it might speed up my processing a bit.

So, before I try to do this myself, I thought to ask here that someone stumble upon the same question and solve it earlier.

Thanks in advance

Answer 1

I used both (OpenCV and FFTW), and you can expect FFTW to be faster than a simpler implementation in OpenCV (how much it depends on your processor and image size, of course). However, if you plan to use your software commercially, FFTW has a rather expensive license ($ 7,500.00). In the commercial case, I would recommend Intel IPP via FFTW, since the performance is similar to a much more affordable license ($ 199.00) + you get a way more functionality in IPP than in FFTW (i.e. only in the FFT library) .

Here are a few tests between FFTW and various competitor libraries. The results are old, but useful for comparison. I'm not sure why IPP 2D transforms are not included (maybe when these benchmarks were met, they didn't exist yet).

If you go the IPP route, you can create OpenCV with IPP support, and it will replace its own cvDFT implementation cvDFT IPP equivalent. This will give you a good performance boost (especially on Intel chips), and you won’t have to change your code.