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

Deconvolution help in PixInsight

astrophotography dslr imaging
  • Please log in to reply
39 replies to this topic

#26 darkstar3d

darkstar3d

    Vostok 1

  • *****
  • Posts: 196
  • Joined: 11 Oct 2013
  • Loc: North-Central Virginia

Posted 20 March 2017 - 03:52 PM

Thanks for that. My problem with deconvolution seems to be in generating a suitable star mask or perhaps I haven't played sufficiently with the sliders yet. I see star masks that have the large and small stars in it and I've been unsuccessful at getting that but I'm still trying. :)



#27 darkstar3d

darkstar3d

    Vostok 1

  • *****
  • Posts: 196
  • Joined: 11 Oct 2013
  • Loc: North-Central Virginia

Posted 20 March 2017 - 05:32 PM

Thanks for that. My problem with deconvolution seems to be in generating a suitable star mask or perhaps I haven't played sufficiently with the sliders yet. I see star masks that have the large and small stars in it and I've been unsuccessful at getting that but I'm still trying. smile.gif

Found some videos and tuts that have defined how to set the mask. I need more practice but my mask look great.



#28 AtmosFearIC

AtmosFearIC

    Messenger

  • -----
  • Posts: 477
  • Joined: 10 Dec 2015
  • Loc: Melbourne

Posted 20 March 2017 - 10:35 PM

When creating star masks I make several of them. After I get the Noise Threshold correct I start with a scale of 2, then 4, 6, & 8.

 

Go into PixelMath and use "Max(S1,S2,S3,S4) and it takes the brightest areas from each star mask. Whether I use a scale of * really depends on the stars in the region though.

 

I do the same for both Linear and Non-Linear data BUT when I move to non-linear I do use the Max function of both the linear and non-linear masks as there are some stars that are better contained in linear than non-linear.


Edited by AtmosFearIC, 20 March 2017 - 10:37 PM.


#29 Jon Rista

Jon Rista

    Hubble

  • *****
  • Posts: 17098
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 20 March 2017 - 10:48 PM

I don't put a whole lot of time into the star masks these days. tongue2.gif I know most articles put a lot of weight on the star mask, and it is an important factor, but I slapped out a simple, basic star mask with default settings except for noise threshold in both my image as well as Jukkas, and left it at that! 

 

I sometimes go through and clone stamp out parts of structure I don't want to be factored into the local derining. Sometimes DSO structure or little bits of noise will end up in the star mask, and I'll blot that out. That tends to be quicker than fiddling and fine tuning the star mask to the nth degree (and worse, for three or four or five different star masks at different scales, which then need to be later combined, which can introduce it's own problems). 

 

The star mask is only used for secondary local deringing. It is not a primary factor in the deconvolution process, and making your star masks more accurate and refined only matters if you end up having to actually fiddle with the local deringing setting anyway. I wouldn't put too much time into the star mask. I would put more time into getting the right regularization and global bright/dark settings, as those matter FAR more to the end result than having a perfect star mask. 


Edited by Jon Rista, 20 March 2017 - 10:48 PM.

  • darkstar3d likes this

#30 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 24 March 2017 - 09:52 AM

Those color artifacts would come from your RGB. Did you activate chrominance noise reduction on the LRGBCombination tool?


Sent from my iPhone using Tapatalk

I did not use the LRGBCombo tool. I created a Luminance using the ChannelExtraction tool. If I was to get the Deconvolution to work, I'm not sure how to apply it to the RGB image since I am applying it to the Lum image?



#31 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 24 March 2017 - 09:56 AM

 

You need to strech the image first and then try deconv.

This is definitely wrong. tongue2.gif For deconvolution to work, you have to work on the original data, in it's linear form, with as little other processing done as possible. In my workflow, deconvolution is the first thing I do after integration, before anything else. 

 

I applied the Deconvolution to my linear Lum created using the ChannelExtraction tool. However, I did apply the DBE process before running the Deconvolution. Could this be causing my problems?



#32 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 24 March 2017 - 10:04 AM

 

I have not yet succeeded in applying deconvolution to my color DSLR images without severe color artifacts. I have tried both working directly with the RGB image and applying deconvolution to the L component only. In the latter case, I still get color artifacts when recombining. All of my tests have been on linear images, as seems to be unambiguously prescribed.

 

You must normalize the color channel weights first before deconvolving RGB data. You do this with the RGBWorkingSpaces tool...just set all the channel weights to 1.0, apply to the RGB image, then deconvolve. You should find that you no longer have artifact issues once you do this.

 

I normalized the color channels using the RGBWorkingSpaces tool, then using the ChannelExtraction tool I created my Luminance image which is what I am running the deconvolution process against. Pixinsight forum response from Juan Conejero indicates the image may not need deconvolution at all, or apply without a mask. I'm pretty happy with the sharp detail in the galaxies - may move onto color processing/noise reduction steps...



#33 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 24 March 2017 - 10:14 AM

 

I don't know if this is a good example. This is noisy image whit not so great stars. But Jon, if you have time could you show me how to deconv. this galaxy image whit more then 1 iteration. I tryed whit custom PSF to linear image and got no result. 

https://drive.google...NFFMXzJyT2d1YWc

I took a deeper look at your data. At first glance, it appears you have both some tilt and coma issues. Another thing about deconvolution is it really only works well if the stars across your field have a consistent PSF. Even if the stars are not ideally round, as long as they are consistent across the frame, that is ultimately what matters. In your case, the stars localized around the central-lower-right area are round, while throughout the rest of the field they are not. You will have to be quite selective about which stars you use to model a dynamic PSF, and even then, because of the inconsistencies, you are likely to find that deconvolution does not work all that well. You might be better off with just using MT to do star reduction, and LHE with a proper star mask to enhance galaxy detail. 

 

Jon, thanks for the detailed explanation - I'll go back and apply the changes based on your suggestions. To confirm; MT is MorphologicalTransformation and LHE is LocalHistogramEqualiztion - correct?



#34 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 24 March 2017 - 11:19 AM

Jon,

I played around with the Deconvolution settings, used the Parametric PSF and applied a scaled approach similar to what you recommended - thanks! 1 more question and 1 warning message I'm not clear what to do with:

-Do I apply the deconvolution process to the RGB image now that it worked on the Lum image?

-I applied the Decon process to the RGB image and when saving the image I received the error message below? Not sure what this means or if, how to correct? Any advice?

 

Writing file:
/Users/mschi99/Astronomy/Astro Images/Leo Triplet/20170218_65deg_800iso_240s_CLS/_Preprocessed/Leo_Triplet_pp_ABE_Decon.fit
** Warning: Deprecated format: FITS
** Warning: Insufficient data storage capabilities: FITS
36 FITS keywords embedded
ICC profile embedded: 'Adobe RGB (1998)', 560 bytes
Writing FITS image: 32-bit floating point, 3 channel(s), 9180x6362 pixels: done


Edited by mike8888, 24 March 2017 - 11:29 AM.


#35 Jon Rista

Jon Rista

    Hubble

  • *****
  • Posts: 17098
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 24 March 2017 - 02:43 PM

Jon,

I played around with the Deconvolution settings, used the Parametric PSF and applied a scaled approach similar to what you recommended - thanks! 1 more question and 1 warning message I'm not clear what to do with:

-Do I apply the deconvolution process to the RGB image now that it worked on the Lum image?

-I applied the Decon process to the RGB image and when saving the image I received the error message below? Not sure what this means or if, how to correct? Any advice?

 

Writing file:
/Users/mschi99/Astronomy/Astro Images/Leo Triplet/20170218_65deg_800iso_240s_CLS/_Preprocessed/Leo_Triplet_pp_ABE_Decon.fit
** Warning: Deprecated format: FITS
** Warning: Insufficient data storage capabilities: FITS
36 FITS keywords embedded
ICC profile embedded: 'Adobe RGB (1998)', 560 bytes
Writing FITS image: 32-bit floating point, 3 channel(s), 9180x6362 pixels: done

Sounds like an image that large (9180x6362) is too large to be stored within a FITS file. You are trying to put over 700MB of data into a FITS container, and I guess that isn't possible. This is the whole reason that XISF was created, because there are limitations with FITS. 



#36 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 24 March 2017 - 07:12 PM

<p>.xisf did the trick. Thanks Jon.&nbsp;</p>

#37 AgilityGuy

AgilityGuy

    Vostok 1

  • -----
  • Posts: 173
  • Joined: 20 Jun 2015
  • Loc: Northern CA

Posted 24 March 2017 - 10:29 PM

Mike, 

 

Deconvolution was always a challenge for me as well. This book provided a lot of insight for using Pixinsight from start to finish.  It's not the resource to end all searching but provides very good footing: https://www.amazon.c...ords=Pixinsight. This book combined with the tutorials in lightvortexastronomy.com have helped me to learn the processes and the background for using them. 

 

JB


Edited by AgilityGuy, 25 March 2017 - 01:25 AM.

  • mike8888 likes this

#38 mike8888

mike8888

    Vostok 1

  • *****
  • topic starter
  • Posts: 111
  • Joined: 19 Mar 2015

Posted 25 March 2017 - 12:07 AM

Mike, 

 

Decomvolution was always a challenge for me as well. This book provided a lot of insight for using Pixinsight from start to finish.  It's not the resource to end all searching but provides very good footing: https://www.amazon.c...ords=Pixinsight. This book combined with the tutorials in lightvortexastronomy.com have helped me to learn the processes and the background for using them. 

 

JB

JB, I'm following the same recipe for learning. Both have been excellent resources as have been the SMEs that contribute to these forums. Thanks all!



#39 Warhen

Warhen

    Vendor (IP4AP)

  • *****
  • Vendors
  • Posts: 1200
  • Joined: 12 Nov 2005
  • Loc: WV, USA

Posted 04 June 2017 - 11:26 AM

John Rista, I was very impressed by your handle on this thread- well done! Two things, I also talk about 'wormy' artifacts with that same term! For the rest here, usually oversharpened small, bright structures (think HII knots of galaxies). I find that about 0.0025 of Global Bright deringing settles this down nicely. Also, curious why you use Poisson rather than default Gaussian for regularization's noise model. Thank you John! And thanks JB for the book shoutout!   


  • Jon Rista likes this

#40 Jon Rista

Jon Rista

    Hubble

  • *****
  • Posts: 17098
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 11 June 2017 - 01:29 PM

John Rista, I was very impressed by your handle on this thread- well done! Two things, I also talk about 'wormy' artifacts with that same term! For the rest here, usually oversharpened small, bright structures (think HII knots of galaxies). I find that about 0.0025 of Global Bright deringing settles this down nicely. Also, curious why you use Poisson rather than default Gaussian for regularization's noise model. Thank you John! And thanks JB for the book shoutout!   

I choose Poisson since that is actually what the noise is. Since the shot noise is derived from counting individual events (the arrival of a photon in the pixel), Poisson statistics are the right statistics to use. The differences between the two are minor, though, and if you have REALLY well exposed subs, Gaussian should be fine. Since I am usually working with NB subs that are just barely swamping read noise by enough, I think Poisson models the noise a bit better.




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.






Also tagged with one or more of these keywords: astrophotography, dslr, imaging



Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics