›

PSF for Deconvolution

John Kmetz
in PixInsight
Hi Adam,

After viewing your videos on Deconvolution from 8/23/21, was wondering why you would use the older long route DynamicPSF instead of the newer script render PSFImage. Other than showing the fuller explanation of how the PSF image was created for demonstration purposes, do you prefer DynamicPSF for some technical reason like better control of variables used? The script is so much easier to use and quite a time saver.

Thanks,

John

Comments

  • BlockHeadBlockHead
    To be honest, I had forgotten about it. But I don't like PSF generated answers at all- they include the wings of the stars (if not adjusted). The automation of the script is fine- but the PSF answer will be the same (please let me know if this is incorrect). That was the point of the demonstration- the parametric version appears to give vastly better results. 

    Regarding using the script- for many systems you will want to window the selection to near the center of the frame (perhaps by using a preview to generate the PSF with the script. This will avoid averaging aberrations near the edge of the field. Now... if Harmut put a radial weighting dependence... that would be cool. 

    -the Blockhead


  • William Fields
    Adam, perhaps I am the blockhead but after watching the deconvolution video (the one with the garbage result) I must confess I am not sure I understand the point you are making.  It seems the garbage result is perhaps a little more harsh looking but it did seem to "split" two small structures  etc.  I just don't think I understand you objection in that it did not seem to "create "  new structures where there were none before.  Hope I am making myself clear, just trying to understand your objection.  

    William
  • John Kmetz
    Ah, thanks Adam. I was running PSFImage on the whole frame, never thought to use a preview where the stars might be of better quality. 
  • BlockHeadBlockHead
    I haven't actually done my suggestion... but I was going with "in principle" thoughts. :)
    This would be a good suggestion for Harmut and his script... to specify a part (central) of the image.

    -the Blockhead
  • craig smith
    Adam,

    To follow onto this thread a bit, you have expressed dissatisfaction with deconvolution results based on a PSF generated from the image data.  You mentioned concern with the calculated PSF in a segment on a different project, although in that case you thought you might need to reduce the star "wings" in the PSF but ultimately decided it was unnecessary.

    I am probably missing something, but the process of generating a synthetic PSF feels a bit arbitrary, as does the adjustment of the wings of the stars for the PSF calculated from the actual data.  

    Should I view this as a typical PixInsight case of experimenting with tweaks to the PSF until the resulting deconvolution results seem pleasing and correct? Or, is there more to it than that?

    BTW, love these targets with the star streams!!!

    Thanks very much and regards,
    Craig
  • BlockHeadBlockHead
    I demonstrated that there exist algorithms that are insensitive or perhaps better optimized for taking the average PSF as measured from the image. The PI algorithm is not one of them. Yes, tweaking is fine- but I demonstrate the difference between good and bad. My "bad" example, in my experience, many people find as good. So tweaking is fine- but again from my experience many users do not know the "goal" they are tweaking towards. That is why I tried to demonstrate In this particular case the tweak of my bad example is bad- even if you find it pleasing- it is not an optimal result. So I guess I am taking a position on that.

    -the Blockhead


  • craig smith
    Thanks very much for the response, but I am still a little confused...

    I think that I appreciate the point you are making about tweaking the measured PSF, and I would be in the group that might struggle to know what I am aiming for were I to choose that method (although I can see and appreciate what you are seeing when you point it out)

    On the other hand, it seemed like the parametric PSF yielded results that were pleasing, probably more accurate that the measured PSF, and compared favorably to the CCD Stack results.  But with that method, it seems to me that there is still an element of "tweaking" in that you can play with size and shape of the PSF...  

    In some sense, both paths seem to require you to know something about where you would like to end up.  Is there something about setting up the parametric PSF that makes it less arbitrary?  I understand that I could be looking at some part of this incorrectly and appreciate any insights.

    Best regards,
    Craig
  • BlockHeadBlockHead
    No. This is a critique of PixInisght. Not all evaluations or convergent tweaks are necessarily possible.
    I will never blindly offer advice just because it is available or part of a particular software program.

    That being said, the parametric solution is measurably better for people who cannot find a good solution using the natural PSF:

    1. There are fewer options. the STD/Shape of the parametric adjustment is quite sensitive- it is easy to find a good solution.(And I think it is because the wings of the PSF are not in the parametric model)
    2. The fact that the global dark compensation is small is indicative of a good solution.
    3. The control of using a different algorithm and reaching a similar (if not better) answer is pretty strong evidence I am on the right track.

    Part of what I hope you find value in, is that I have experimented with lots of settings and I bring you my conclusions. You are of course free to (and I encouraged to do so) perform experiments yourself and see if you come to the same conclusions... OR- find something better (which would DELIGHT me).

    I feel I have given you in this video enough information to do the experiments yourself or simply accept my conclusions as a convenience (with the caveat I am not always right- but I have a good track record). It is my hope you can do the tweaking on your own now...because you understand the basic idea of what is going on and what to look for. Please don't assume this is demonstrated elsewhere or commonly known.

    -the Blockhead
  • craig smith
    Fair enough.  

    I very much do find value in your approach to presenting information and I appreciate you taking the time to clarify this situation

    Best regards,
    Craig
Sign In or Register to comment.