0
$\begingroup$

I have pictures, taken with a microscope, of biological samples that exhibit a periodic pattern. Previously, I have used ImageJ to manually rotate the image perpendicular to the pattern and plot its grey value profile. Doing FFT in MATLAB I can determine the distance between two peaks of that pattern via its spatial frequency. But this all depends on how (well) I rotate the sample. So my goal is to actually use fft2 in MATLAB to directly transform the image regardless of sample angle. Somewhere I have read about calculating the rotational average (after fftshift) of the resulting spectrum, but I don't know how.

  • Is this the right approach anyway?
  • Or is there an easier way to calculate the (average) distance between two units of a periodic pattern in an image without having to adjust the angle beforehand?
$\endgroup$
  • $\begingroup$ Would it be possible to post a representative sample of the image? $\endgroup$ – A_A Dec 23 '16 at 8:00
1
$\begingroup$

i am not an image processing guy. but i know what the 2-dimensional Discrete Fourier Transform is.

normally, in two dimensions, the way we measure period is with an operation related to the autocorrelation and you can get to the autocorrelation $R_x[k,\ell]$ with the DFT and magnitude-squaring the result $|X[m,n]|^2$ and inverse DFT to get the autocorrelation. mathematically, you should have half of the 2-dim buffer (like quadrants II and IV) set to zero, to prevent overlap in the correlation, then you have a triangular envelope on the peak values, even if it were a perfectly periodic function.

so, in the $k$ dimension of the result there will be an apparent period, call it $p_k$ and similar for the $\ell$ dimension. the period will be $p = \sqrt{p_k^2 + p_\ell^2}$ . actually, when viewing $R_x[k,\ell]$ as a surface, there should be a circular ridge of radius $p$.

$\endgroup$

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.