fft2 and fftn provide the axis parameter which allows to specify over which axes to perform the transform. Does the method obey order?
Given a 2-dimensional numpy array
a = np.array([[1,2],[3,4]])
is
fft2(a,axis=(0,1) different from
fft2(a,axis=(1,0) I have data which is correlated across lines but not columns...
The DFT is separable, which means calculating the 2D DFT is the same as calculating two individual 1D DFTs along each axis:
import numpy as np a = np.arange(25).reshape(5, 5) out1 = np.fft.fft2(a) tmp = np.fft.fft(a, axis=0) out2 = np.fft.fft(tmp, axis=1) np.allclose(out1, out2) # True And since the nested summations of the DFTs commute, the order of operations does not matter
tmp = np.fft.fft(a, axis=1) out3 = np.fft.fft(tmp, axis=0) np.allclose(out1, out3) # True Which means the order of the axes parameter does not matter either
out4 = np.fft.fft2(a, axes=(1, 0)) np.allclose(out1, out4) # True 