I have the following rather simple problem and unfortunately I am not getting forward.

Imagine a simple 2D image with pixels and a unique value for each pixel of the image. For example, let the image be 512x512 pixels. The size of the image is given by 10 nm x 10 nm. Since I want to view the image in frequency space, I calculate the Fourier transform of the image.

Of course, the image still has 512x512 pixels, but what about the units? I would say the physical units are now 1/m, but what happens to the 10 nm?

My understanding is that 1 pixel in local space would be 10/512 nm for this example and thus 1 pixel would be 1/(10nm*512) in frequency space. Is this correct?


1 Answer 1


You are close. In the same way that a time domain signal with units seconds becomes Hertz (i.e. 1/seconds) with the Fourier transform, a spatial domain signal with units meters becomes wave number (i.e. 1/meters). The concept of frequency does not change. In both units, you can think of it as cycles per unit length of time/space. Recall that the resolution in frequency space is dependent on the length of your signal in time/space (i.e. 10 nm) and not the number of points. So for the first non-DC DFT bin, you have units of 1 cycle / 10 nm (a wave number of 0.1 nm$^{-1}$). For the second bin, it would be 2 cycles / 10 nm (a wave number of 0.2 nm$^{-1}$). Because you have a resolution of 512 pixels along each dimension, you can measure up to the Nyquist frequency, or one half the sampling rate or 256 cycles / 10 nm (25.6 nm$^{-1}$).

  • $\begingroup$ I understand, thank you for your quick reply. Now in my example I have a central circle with a radius of 20 pixels. Then the length in frequency space would be given by 20/(10nm*512) or by 20/10nm? I cannot fully understand what you mean by independent of the number of points. $\endgroup$
    – user61315
    Feb 1 at 14:12
  • $\begingroup$ @user61315, If your DFT has a circle with radius of 20 pixels, then you have a frequency of 20 cycles per 10 nm, or a wave number of 2 nm$^{-1}$. For the independence of number of points, consider a DFT of a time signal with sample rate $fs$ and duration $T$. The frequency resolution will be $\Delta f=1/T$ and the Nyquist frequency is $fs/2$. Note that if you increased the sample rate but kept the same duration, the number of points in your measurement, $N=T*fs$, increases but $\Delta f$ does not change. $\endgroup$
    – Ash
    Feb 2 at 17:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.