1
$\begingroup$

I am aware that I could use deconvolution algorithms such as LUCY, Weiner and Blind but I am confused as to how to estimate the PSF (point spread function), which is needed for the LUCY and Weiner algorithms. I am currently using the Widefield Florescence Microscopy setup.

I have read in a lot of places around the internet that a fluorescent bead that represents a sub resolution object and therefore a impulse function. While I understand that the OTF (optical transfer function) is represented by the image I am getting, how am I superposed to deconvolve the image I have using the very same image?

Here is an example of the image I am trying to deconvolve:

I have a program which automatically singles in on the bead and crops the rest of the image. All I need to do is apply deconvolution to this.

My ultimate aim to achieve optical sectioning by deconvolving a bunch of 2D slices I got from a video and representing them with a 3D software.

So can someone please explain what I need to do, either theoretically (or in terms of MATLAB, which would be preferred).


One Possible Solution

After talking to a few people, I have noticed that EPFL have a Java resource that is callable from MATLAB. It's a PSF generator, based on the specifications of your microscope. This generated PSF can be used with the LUCY and Weiner.


Alternative Solution

I have also found this resource, which is a MATLAB library dedicated to what I am working on right now.


This (I would love it, if someone attempts to answer this question too) is how I will measure the performance of the deconvolution algorithms. I believe that the ultimate result of my deconvolution will depend on the PSF estimation (EPFL resource or Praveen's algorithm) and the deconvolution algorithm (Blind, LUCY, Weiner etc.). I will post a table of all the results here once this is done.

$\endgroup$
  • 1
    $\begingroup$ Crossposted to physics.stackexchange.com/q/232098/2451 and dsp.stackexchange.com/q/28510 $\endgroup$ – Qmechanic Jan 27 '16 at 15:55
  • $\begingroup$ @Qmechanic crossposted to SO at stackoverflow.com/questions/35041093 $\endgroup$ – Sharan Duggirala Jan 27 '16 at 15:57
  • 1
    $\begingroup$ Your problem goes under the name "blind deconvolution". It's a pretty hard problem but there are methods out there that you can try. $\endgroup$ – Dirk Jan 27 '16 at 20:47
  • $\begingroup$ @Dirk I have already tried blind deconvolution, but it does not give the whole picture about the best possible deconvolution method with the microscope I am working with. I have added a bit to the question, if you would like to see. $\endgroup$ – Sharan Duggirala Jan 28 '16 at 4:42
1
$\begingroup$

I have read in a lot of places around the internet that a fluorescent bead that represents a sub resolution object and therefore a impulse function. While I understand that the OTF (optical transfer function) is represented by the image I am getting, how am I superposed to deconvolve the image I have using the very same image?

This is what we do for two-photon microscopes. Is your bead much smaller than the point spread function? If so then, as you mention, you can approximate the bead an impulse. From this perspective, the bead itself is a delta function, and the recorded image of it is your estimate the PSF. Deconvolving the image of the bead with itself would give a delta function at the bead's location, which is exactly the assumption behind this method. It's important to average several frames to reduce the noise, subtract the background, normalize, and crop the image to remove unnecessary background. You could also fit a parametric model of the PSF (e.g. a Gaussian to approximate the central lobe, or a more complicated model). Here, you would again consider the bead to be a delta function, and the model would be fit to the image of the bead. If your bead is larger and you don't want to treat is as a delta function, I could imagine treating it as a known sphere (if you know the size). In that case, you'd have to do a deconvolution to recover the PSF.

$\endgroup$

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.