Re: FFT implementation in ImageJ

Posted by Michael Schmid on Jul 09, 2014; 4:48pm
URL: http://imagej.273.s1.nabble.com/FFT-implementation-in-ImageJ-tp5008239p5008653.html

Hi everyone,

in the recent daily builds, Wayne has added two convenient ways to access the built-in 1D FHT of ImageJ.

   FHT.transform1D(float[] x)
   FHT.inverseTransform1D(float[] fht)

The FHT (Fast Hartley transform) is closely related to the FFT, and it also needs array sizes that are a power of 2.
http://en.wikipedia.org/wiki/Discrete_Hartley_transform

Actually, it was possible to use the built-in FHT already in previous versions of ImageJ, but that was not so obvious. In the daily build, the 1D FHT also works with array sizes >2^16 (for very large arrays, note that the accuracy is limited to maybe 5 digits or so because it's all 32-bit float, not double-precision arithmetics).

From the FHT, one can easily determine the Fourier (FFT) coefficients for either sine and cosine, or the coefficients of the complex FFT.  Of course, you can also get the power spectrum, see the API documentation of FHT.transform1D.

Michael

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html