Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Help Wanted with Deconvolution of M101 Image w/ C11

astrophotography cassegrain ccd Celestron dso imaging SCT
  • Please log in to reply
10 replies to this topic

#1 BenKolt

BenKolt

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1881
  • Joined: 13 Mar 2013

Posted 18 May 2019 - 12:53 AM

Greetings!

 

I am working on data acquired recently of M101 and am having some trouble with Deconvolution in PI.  I would appreciate advice on how to not transform some of my hard-earned detail of the galaxy's structure into little bright pinpoints that ultimately look like little stars that oversaturate the intensity and color.

 

Hardware Details

  • Celestron 11 EDGE
  • Celestron 0.7x Focal Reducer
  • Innovations Foresight ONAG
  • QSI 683 ws-8
  • Astrodon LRGB Series 2 Filters

Acquisition Details

  • Acquisition with Sequence Generator Pro
  • Guiding with PHD2 - achieved typically 0.4" RMS guiding curves with star FWHM around 2" (that's quite decent for my skies)
  • L : 120 x 300s = 10h : 1x1 : 0.568 "/px
  • R : 60 x 300s = 5h : 2x2 : 1.138 "/px
  • G : 60 x 300s = 5h : 2x2
  • B : 60 x 300s = 5h : 2x2

Brief and Abbreviated PixInsight Processing Details (L Channel)

  • ImageCalibration
  • CosmeticCorrection
  • StarAlignment (M101 barely fits in the FOV even with the focal reducer!)
  • ImageIntegration
  • DynamicBackgroundExtraction - sparse control points around FOV perimeter
  • MureDenoise
  • Deconvolution

Deconvolution has always been one of my most challenging processes to master in PixInsight.  I wish to achieve a nice balance between recovering reasonable detail and not overdoing it to create detail that actually isn't there.  (In other words, it is my desire to use the Deconvolution responsibly.)  I ended up with quite a bit of integration time on M101 as I chose to image over several nights with variable transparency and seeing conditions.  Despite the fluctuating weather and conditions I actually ended up with as much as 25h of adequate frames through the LRGB channels.  (In fact, it is my wish now to go through the processing again and be a bit more selective as my finished product has several deficiencies besides the deconvolution.)

 

Lately I worked my way through David Ault's talk on Deconvolution in PixInsight on the Astro Imaging Channel on YouTube.  There were some helpful bits in there that I had not considered before, mainly having to do with the generation of a suitable Contrast Mask for the image rather than a more straightforward Intensity Mask.  This is my first attempt at making such a mask, and things came together better than usual, however I am still affected by the problem of transforming some of the galaxy detail's (small and relatively bright dust and gas regions) into bright "stars" that aren't really there.  I would appreciate your comments and your advice on how to avoid or at least minimize this problem.

 

I will post some images now for you to see what I am talking about.

 

Thank you for your attention and help.

 

Best Regards,

Ben

 

First, here is my first image all the way through the LRGB processing.  (I don't like my colors, I pushed the black point too much, and I wish to work on all that again, but my main reason for posting this evening is to ask about Deconvolution.)

 

M101_LRGB_089-090_0300s_20190425_-25C.v004B_CT2-1.jpg

 

Here is a cropped region around the core.  Note the little white pinpricks following the arms and trails of the galaxy.  These are not present in the original combined channels but rather result from the Deconvolution step.

 

M101_LRGB_089-090_0300s_20190425_-25C.v005A_CROP1-1.jpg

 

Here is the same cropped region from my linear Luminance image after the MureDenoise step and just before the Deconvolution step.  I have tried to stretch it to match the final LRGB image, but that's tough to do at this point, however I think you can still compare the pre-deconvolved regions below with the post-deconvolved regions above.  Putting these two images down together actually makes the problem not seem quite so bad, but I am still interested in learning how to minimize this effect.

 

M101_L_090_1x1_0300s_20190430_-25C.v003B_MDN_STRETCH-1.jpg

 

Last, here is the same cropped region from my linear Luminance image after the Deconvolution step.  I've matched the stretch to the last one above.

 

M101_L_090_1x1_0300s_20190430_-25C.v004F_Deconv_STRETCH-1.jpg

 

Thank you again, and I look forward to hearing your comments and suggestions.

 

 

 

 

 

 

 

 


  • elmiko, mumbles, kathyastro and 1 other like this

#2 BKMaynard

BKMaynard

    Ranger 4

  • *****
  • Posts: 317
  • Joined: 21 Feb 2013
  • Loc: Dupo, Il. (st.louis area)

Posted 18 May 2019 - 02:18 AM

I think you've done a good job. Decon has simply sharpened the smaller stars. You may try increasing the global white a bit but I think it looks good


  • elmiko likes this

#3 darkstar3d

darkstar3d

    Messenger

  • *****
  • Posts: 434
  • Joined: 11 Oct 2013
  • Loc: North Metro Atlanta

Posted 18 May 2019 - 04:50 AM

You can mask out the stars. I saw a presentation somewhere where two decon passes were used. One for stars and one for the nebula.


Sent from my iPhone using Tapatalk

#4 kellyvictoria

kellyvictoria

    Ranger 4

  • *****
  • Posts: 334
  • Joined: 16 Apr 2017
  • Loc: Florida, 29 degrees N

Posted 18 May 2019 - 06:50 AM

That is just lovely. I liked the very black background on the first one..

vk


  • elmiko likes this

#5 lucam

lucam

    Viking 1

  • -----
  • Posts: 511
  • Joined: 09 Oct 2017
  • Loc: Upstate NY, USA

Posted 18 May 2019 - 07:20 AM

Ben, that looks already like a spectacular image. 

 

I don't think that's an artifact. The deconvolution process is doing exactly what it is supposed to. Remember, deconvolution is deblurring, not sharpening. It is meant to model the atmospheric and optical aberrations to estimate what the image looked like in their absence. 

 

Remember the spherical aberration in the original Hubble mirror? You could say that the sharp high-frequency detail in later iterations were not there in the original image.

 

Hubble.jpg

 

You may also want to take  a look at this paper: https://ws680.nist.g...m?pub_id=150419. It's not  the iterative Richardson-Lucy algorithm for deconvolution and perhaps pushes the results beyond the reasonable for aesthetic imaging, but it shows the point of image estimation with deconvolution algorithms to recover lost high-frequency  detail. 


Edited by lucam, 18 May 2019 - 09:22 AM.

  • elmiko likes this

#6 elmiko

elmiko

    Mercury-Atlas

  • -----
  • Posts: 2987
  • Joined: 27 Jul 2009
  • Loc: Arizona

Posted 18 May 2019 - 07:47 AM

I think you've done a good job. Decon has simply sharpened the smaller stars. You may try increasing the global white a bit but I think it looks good

I totally agree with the Post above. Outstanding image!



#7 elmiko

elmiko

    Mercury-Atlas

  • -----
  • Posts: 2987
  • Joined: 27 Jul 2009
  • Loc: Arizona

Posted 18 May 2019 - 07:49 AM

Ben, that looks already like a spectacular image.

 

I don't think that's an artifact. The deconvolution process is doing exactly what it is supposed to. Remember, deconvolution is deblurrying, not sharpening. It is meant to model the atmospheric and optical aberrations to estimate what the image looked like in their absence.

 

Steve, I agree with Lucam's post here. I think those are sharpened stars, not artifacts.

Keep up the great work, Mike



#8 BenKolt

BenKolt

    Surveyor 1

  • -----
  • topic starter
  • Posts: 1881
  • Joined: 13 Mar 2013

Posted 18 May 2019 - 08:48 AM

I think you've done a good job. Decon has simply sharpened the smaller stars. You may try increasing the global white a bit but I think it looks good

Thank you!  The Global white setting is the one I tweak the most, and I'll sweep the values even wider next time to monitor the effect.

 

 

You can mask out the stars. I saw a presentation somewhere where two decon passes were used. One for stars and one for the nebula.

That's an excellent idea!  Thank you!

 

 

That is just lovely. I liked the very black background on the first one..

vk

Thank you, vk!  I really think that this data can tolerate less black, however.  M101 has two very faint arms on the right side of my image that I actually resolved, but I pushed them down into the background in this version.  Next time I'll try to draw them out better if possible.

 

 

Ben, that looks already like a spectacular image. 

 

I don't think that's an artifact. The deconvolution process is doing exactly what it is supposed to. Remember, deconvolution is deblurrying, not sharpening. It is meant to model the atmospheric and optical aberrations to estimate what the image looked like in their absence. 

 

Remember the spherical aberration in the original Hubble mirror? You could say that the sharp high-frequency detail in later iterations were not there in the original image.

 

You may also want to take  a look at this paper: https://ws680.nist.g...m?pub_id=150419. It's not  the iterative Richardson-Lucy algorithm for deconvolution and perhaps pushes the results beyond the reasonable for aesthetic imaging, but it shows the point of image estimation with deconvolution algorithms to recover lost high-frequency  detail. 

Thank you, lucam!  Right!  I did know physically what Deconvolution is doing, but I was loose in my language.  I'll try to say "deblur" next time.  I like this process because of its physics-based steps of measuring and applying the PSF in order to unfold some of the atmospheric and optical effects.  This is the action I said in my original post that I want to use responsibly.  And thank you for the paper - I'll give it a look.  That's the kind of information for which I was asking.

 

 

I totally agree with the Post above. Outstanding image!

 

Ben, that looks already like a spectacular image.

 

I don't think that's an artifact. The deconvolution process is doing exactly what it is supposed to. Remember, deconvolution is deblurrying, not sharpening. It is meant to model the atmospheric and optical aberrations to estimate what the image looked like in their absence.

 

Steve, I agree with Lucam's post here. I think those are sharpened stars, not artifacts.

Keep up the great work, Mike

Thank you, elmiko!

 

"Deblur!" "Deblur!" "Deblur!"  Maybe if I keep typing it here I'll remember next time!

 

After posting the before and after pictures I did notice that I wasn't quite so discouraged with the outcome.  I was able to correlate that many of those little consolidated pinpricks actually are stars, and maybe what I really need to do is work out how to lessen how much deblurring they receive.  Working on them separately from the galaxy details is intriguing.

 

You've all made some excellent suggestions, much to ponder and try for myself.  Thank you for the help and encouragement.

 

Best Regards,

Ben



#9 kathyastro

kathyastro

    Viking 1

  • *****
  • Posts: 572
  • Joined: 23 Dec 2016
  • Loc: Nova Scotia

Posted 18 May 2019 - 09:14 AM

I don't care whether you call it sharpening or deblurring, but either way, I agree with those who say that it is doing what it is supposed to do.  Those points of light are really there, and Deconvolution is bringing them out.

 

That is an awesome image!

 

If you'd rather suppress the bright stars in the galaxy, you would need to run some other process to do so.



#10 lucam

lucam

    Viking 1

  • -----
  • Posts: 511
  • Joined: 09 Oct 2017
  • Loc: Upstate NY, USA

Posted 18 May 2019 - 09:21 AM

Brief and Abbreviated PixInsight Processing Details (L Channel)

  • ImageCalibration
  • CosmeticCorrection
  • StarAlignment (M101 barely fits in the FOV even with the focal reducer!)
  • ImageIntegration
  • DynamicBackgroundExtraction - sparse control points around FOV perimeter
  • MureDenoise
  • Deconvolution

 

Ben, one more thing. In your workflow above you mention that you run MureDenoise after DBE. My understanding is that MureDenoise works best with data straight out of the sensor. DBE isn't supposed to change high-frequency detail in the images and so you may be ok. However, I see no harm in following the recommended workflow of running MureDenoise immediately after ImageIntegration. 



#11 mumbles

mumbles

    Mariner 2

  • -----
  • Posts: 256
  • Joined: 30 Sep 2011
  • Loc: Adirondacks

Posted 18 May 2019 - 11:12 PM

I love this. I've actually been going backn n forth with the same image I took with my 8in edgehd. I'm now learning deconvolution and hope to achieve details like you have. Obviously not as much because you have the larger scope but anyway. Good job!

Sent from my SM-G950U using Tapatalk


CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics





Also tagged with one or more of these keywords: astrophotography, cassegrain, ccd, Celestron, dso, imaging, SCT



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics