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

Problems with Dynamic Background Extraction

  • Please log in to reply
30 replies to this topic

#1 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 09:17 AM

Every Pixinsight recommended workflow includes DBE/ABE as an initial step (after crop, still linear), but is it always helpful?  It seems to make my images far worse, leaving defined gradients/streaks where background extractions were applied.  I've spent many hours experimenting with every possible setting/amount and location of sample points, etc,  but I always get poor results.

 

I usually don't have gradient issues to begin with, so is background extraction even necessary?  Attached is an example of the effect I typically see (shown is a calibrated/stacked/cropped lum image).  Am I doing something completely wrong?  Setting anywhere from 5-500 background sample points I get similar effects... subtraction... regardless of normalize or any other setting.

 

I've found little/no official literature about DBE, so I'm not sure what exactly it's supposed to be correcting.  I image from Bortle 5 - ~mag 20 skies... so it would be great if it magically cleared a layer of LP, but I'm not seeing any benefits from it.  

 

 

 

 

Attached Thumbnails

  • 1.jpg
  • 2.JPG
  • 3.jpg


#2 terry59

terry59

    Cosmos

  • *****
  • Posts: 8112
  • Joined: 18 Jul 2011
  • Loc: Colorado, USA

Posted 21 September 2018 - 09:26 AM

Try ABE with the function degree set to 2



#3 sharkmelley

sharkmelley

    Surveyor 1

  • -----
  • Posts: 1821
  • Joined: 19 Feb 2013

Posted 21 September 2018 - 09:30 AM

I recommend using ABE with function degree set to 1 - this is guaranteed not to introduce new gradients..  After that, only use higher function in ABE and only use DBE if you can actually see gradients that need removal.

 

Mark


Edited by sharkmelley, 21 September 2018 - 09:30 AM.


#4 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 09:32 AM

Try ABE with the function degree set to 2

Here's what that produces... I've tried hundreds of settings variations- no luck.

Attached Thumbnails

  • 4.JPG
  • 6.jpg


#5 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 09:36 AM

I recommend using ABE with function degree set to 1 - this is guaranteed not to introduce new gradients..  After that, only use higher function in ABE and only use DBE if you can actually see gradients that need removal.

 

Mark

Here it is set to 1... Much better, but it still seems to have produced a gradient.  Is this an improvement somehow from my original image?

Attached Thumbnails

  • 7.JPG
  • 8.jpg

Edited by chadrian84, 21 September 2018 - 09:38 AM.


#6 Bretw01

Bretw01

    Mariner 2

  • -----
  • Posts: 295
  • Joined: 13 Feb 2017
  • Loc: IL USA

Posted 21 September 2018 - 09:43 AM

When using DBE make sure no background samples are on top of a star or galaxy/nebulosity. 500 sample points are probably way too many, I find that placing them manually where I think it is the true background works for me. I end up with around 12 sample points.

 

I agree when using ABE function degree set to 1 a good starting point, and also try setting tolerance and deviation to .500.



#7 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 09:52 AM

I'm not sure what benefit I'm supposed to see from this process.  Am I just as well off (better off?) skipping it?  @Bret - I know the fewer sample points the better - preferably under 20... 500 was just something I've experimented with and found it also doesn't work.  I've tried hundreds of combinations of tolerence, other settings, etc... and never see an improvement.



#8 spokeshave

spokeshave

    Surveyor 1

  • *****
  • Posts: 1556
  • Joined: 08 Apr 2015

Posted 21 September 2018 - 01:30 PM

If you don't have troublesome gradients or vignetting, then there is no need for DBE/ABE. Your background appears to be fairly flat, with only minor vignetting, so you may not get much benefit. On the other hand, there appears to be some hints of IFN in your image and background extraction can be helpful in pulling that out. 

 

I'm not sure you are having a problem with your background extraction. PI uses an 8-bit lookup table for screen displays by default. That's why your background images look so posterized. That's just a display artifact and is not contained in the underlying image. Before abandoning the DBE/ABE processes, see what they look like after enabling the 24-bit LUT (click on the "24" icon). 

 

Tim


  • pfile, jdupton and sharkmelley like this

#9 dhaval

dhaval

    Vendor

  • *****
  • Vendors
  • Posts: 1101
  • Joined: 21 Jul 2008
  • Loc: Round Rock, TX

Posted 21 September 2018 - 02:02 PM

I typically use very few DBE sample points - for you, it seems to me that you might need a few (no more than 5 I would imagine) at the top right corner and bottom right corner (I see some vignetting artifiacts - also, do look at your master flat again to see if you can improve calibration - only very slightly, definitely nothing major). Once you are done coming up with sampling points, don't apply any correction and run the process as is - this will give you a view of the gradient prior to DBE application. From here on in, you can either keep the samples, add more or remove as needed.

 

CS!



#10 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 21193
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 21 September 2018 - 02:17 PM

Read this...it should help you get a firm grasp on how DBE works:

 

https://jonrista.com...oundextraction/


  • okiedrifter and sharkmelley like this

#11 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 04:48 PM

I'm not sure you are having a problem with your background extraction. PI uses an 8-bit lookup table for screen displays by default. That's why your background images look so posterized. That's just a display artifact and is not contained in the underlying image. Before abandoning the DBE/ABE processes, see what they look like after enabling the 24-bit LUT (click on the "24" icon). 

 

Interesting.  "24" becomes disabled as soon as a BE is executed.  Clicking "24" after execution changes the extracted background picture into a nice smooth background (see before/after below) - but the target (new galaxy) image still inherits the sharp gradients from the "before" background.  Nothing happens to the target image when 24 is clicked, or when it's stretched/saved, etc.  Are you saying the harsh gradients are not actually in the new target (galaxy) image?  How would I view/save the "underlying image" without the harsh gradient?

 

It seems the background extraction may actually be helpful if it would apply the smooth version of the extraction to the new image.

Attached Thumbnails

  • Before clicking 24.JPG
  • After Clicking 24.JPG
  • 24.jpg


#12 spokeshave

spokeshave

    Surveyor 1

  • *****
  • Posts: 1556
  • Joined: 08 Apr 2015

Posted 21 September 2018 - 04:50 PM

Interesting. Are you re-applying the STF after the BE? If not, try it - with the 24-bit LUT enabled.

Tim

#13 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 05:53 PM

Yeah, I've tried a lot of different things with no luck.  I'm abandoning DBE/ABE for now - I get the same problem with all of my images/targets.

 

While playing around with Pixel Math I came up with this image... I call it "DBE Twilight Zone"

Attached Thumbnails

  • test4.jpg

  • happylimpet likes this

#14 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 21193
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 21 September 2018 - 06:38 PM

 

 

It seems the background extraction may actually be helpful if it would apply the smooth version of the extraction to the new image.

It actually does smooth it. There is a setting for it too. You should check out the link I shared earlier, it covers everything you need to know. 



#15 pfile

pfile

    Aurora

  • -----
  • Posts: 4590
  • Joined: 14 Jun 2009

Posted 21 September 2018 - 08:05 PM

i feel like you must have samples on the galaxy itself with that circular gradient in the result.

 

can you post a screenshot of the samples placed on the image?

 

rob



#16 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 09:53 PM

i feel like you must have samples on the galaxy itself with that circular gradient in the result.

 

can you post a screenshot of the samples placed on the image?

 

rob

The examples above used default settings with ABE... I couldn't compress the ABE sample shot below 500kb, but the galaxy was not touched.

 

Here's a typical example using DBE... see next post for results.

Attached Thumbnails

  • night 1 1.jpg


#17 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 09:58 PM

Here's the results using the same data points from above... default settings in DBE, .5 tolerance, 3.0 shadows, etc- subtraction, no normalize... 

 

@Jon... I'm currently reading through your DBE tutorial... I've skimmed through it before but will carefully read through it this time.

Attached Thumbnails

  • Night 2.JPG
  • Night 333.jpg


#18 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 21193
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 21 September 2018 - 10:00 PM

Here's the results using the same data points from above... default settings in DBE, .5 tolerance, 3.0 shadows, etc- subtraction, no normalize... 

 

@Jon... I'm currently reading through your DBE tutorial... I've skimmed through it before but will carefully read through it this time.

There is definitely something strange about your gradients. They appear to be very low bit. This could be caused by not using the 24-bit rendering setting in PI, however since the low-bit effect seems to show up in your corrected images, I am inclined to think it is real. Can you take a screenshot of the settings you are currently using in DBE?



#19 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 10:31 PM

I think all pictures I've posted here used default settings, as shown below:

Attached Thumbnails

  • settings2.jpg


#20 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 21 September 2018 - 11:02 PM

Here's a dropbox file for my master Lum (shown above) if anyone cares to take a look - this was about 13.5 hours of oversampled, underexposed lum data on the Deer Lick Group (27 hours total integration time)... f10/ 8" Edge with ASI1600, Gain 75, offset 12, x500 90 second subs  (my Edge .7x reducer arrived today smile.gif along with an ASI183mm pro for my refractor).  

 

https://www.dropbox....nal_r.xisf?dl=0

 

 

Here are links to the RGB masters (I get the same DBE problems with RGB).. about 13.5 hours total between the 3 filters, 5 minute subs:

 

Red:   https://www.dropbox....nal_r.xisf?dl=0

 

Green:   https://www.dropbox....nal_r.xisf?dl=0

 

Blue: https://www.dropbox....nal_r.xisf?dl=0


Edited by chadrian84, 22 September 2018 - 08:52 AM.


#21 Bretw01

Bretw01

    Mariner 2

  • -----
  • Posts: 295
  • Joined: 13 Feb 2017
  • Loc: IL USA

Posted 22 September 2018 - 12:41 AM

I think it is a setting on your computer/monitor. The file you shared "DSS_Lum_Final_r" calibrated fine for me with ABE and DBE. 

 

Used 12 sample points, needs cropped to clean up edges. Looks like some nice data!

 

Screen-Shot-2018-09-22-at-12.36.28-AM.jpg

 


  • gunny01 likes this

#22 chadrian84

chadrian84

    Explorer 1

  • -----
  • topic starter
  • Posts: 67
  • Joined: 03 Jun 2018
  • Loc: Wisconsin

Posted 22 September 2018 - 01:06 AM

Thanks Bret. Can you post a picture of the background extraction (the part that was removed)? Maybe just an ABE with default settings.  I first cropped 120 pixels from all sides in the pictures above.


Edited by chadrian84, 22 September 2018 - 01:13 AM.


#23 Peter Zbib

Peter Zbib

    Ranger 4

  • -----
  • Posts: 358
  • Joined: 20 Jun 2014

Posted 22 September 2018 - 06:41 AM

Hum

 

I gave it a try with ABE (on "DSS_Lum_Final_r"). Did not work. I get the same effect as in the original message of chadrian84.

 

Surprised. Most of the times ABE works just fine. Sometimes I have to use DBE and customize. Which seems to be the case here: if you dont stretch the image and generate points in DBE you will see that with default settings very few points are selected by default.

 

I dont have the time now, but this clearly need manual choice of background points (eventually raising tolerance, etc).

 

Unfortunately, this is interesting ...

 

=============

 

UPDATE:

 

DBE works

 

I unstretched the image to see the zones where the default DBE generated points were. I then added a bunch of points (mostly around the corners)

 

DBE then works like a charm and I obtain basically the same result as Bret. (slight difference due to the fact that I did not bother cropping the side (which is a must))


Edited by Peter Zbib, 22 September 2018 - 08:51 AM.


#24 ChrisWhite

ChrisWhite

    Soyuz

  • *****
  • Posts: 3805
  • Joined: 28 Feb 2015
  • Loc: Colchester, VT

Posted 22 September 2018 - 09:02 AM

This may or may not be related...

 

I have noticed some apparent quantization error introduced by the computer monitor at certain zoom settings.  If you zoom in or out, do the rings disappear?  If so, they might not be real, but just how it is being displayed on your monitor.



#25 Bretw01

Bretw01

    Mariner 2

  • -----
  • Posts: 295
  • Joined: 13 Feb 2017
  • Loc: IL USA

Posted 22 September 2018 - 11:04 AM

Thanks Bret. Can you post a picture of the background extraction (the part that was removed)? Maybe just an ABE with default settings.  I first cropped 120 pixels from all sides in the pictures above.

I just tried ABE with defaults ( I didn't use the defaults last night) and got the same results as you did, so I don't think it is related to your monitor.

 

Edit: I have been playing around with ABE this morning and cannot get anything without a gradient. I think I was just tired last night and didn't notice it.


Edited by Bretw01, 22 September 2018 - 11:13 AM.



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






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics