It seems that most deconvolution algorithms mainly handle the main lobe of a point spread function (PSF) and assume that sidelobes can be safely neglected.

For a direct algorithm trying to perform a division in frequency-domain, it seems difficult to deal with the zero crossings of the Fourier-transformed PSF.

For the Richardson-Lucy algorithm, there still seems to be a risk of division by a very small number for the small weights in-between the sidelobes, but I'm not sure. I suspect some kind of ad hoc thresholding can help with this.

Is there any general insight that's common knowledge for this problem?

  • $\begingroup$ I wouldn't agree to your first sentence. If that was the case, deconvolution would be a much easier problem, because we'd be looking for concave functions only. $\endgroup$ Commented Feb 21, 2021 at 17:37
  • $\begingroup$ @MarcusMüller I have searched but have not found many specific references to sidelobes. Do you have any suggestion for a good reference? $\endgroup$
    – Orhym
    Commented Feb 21, 2021 at 18:22
  • $\begingroup$ I don't really know; all the material I can think of optimize some global target function, and none says "forget about the sidelobes". $\endgroup$ Commented Feb 21, 2021 at 23:12


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.