python - 2D fft - Null imaginary part -

i integrating python routine c++ code.

there computation of fft 2d of real matrix, using in python

f_blk=np.fft.fft2(blk)

f_blk complex 512*24 matrix, coefficients complex real , imag part order of magnitude 10e5.

when compute fft2 of matrix in c++ complex matrix complex coefficients with real part order of magnitude 10e6 , null imaginary part.

if use real input fft
- then not need have calculations ( c += r * c simpler c += c * c)
- so if have fft c = f(r) coded such input data faster
- then standard fft c= f(c)
- and not need have allocated memory input imaginary part
- also when input data real fft output symmetric
- so can compute first half of output data , mirror rest
the magnitude difference
- either have wrong implementation of fft (in python or in c++)
- or have different normalization coefficients
- plot data , compare if difference constant scale factor
- if not have bug in fft implementation somewhere or python fft not fft
- also not forget data size fft must power of 2
- if implementation expects cause of error
- so try resize matrix 512x24 512x32 0 padding before fft
- another thing can cause can overflow errors
- if mix huge , small numbers accuracy lost
- especially fft recursions output magnitude can 10e5 subresults can much bigger !!!
2d fft
- look here 2d fft,dct 1d fft,dct
- it contains slow 1d dft,idft implementations in c++ (r->c,c->r)
- and algorithm how compute 2d transforms it
- with correct results can check yours

O9