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

Why can't I get DBE to work?

  • Please log in to reply
29 replies to this topic

#1 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 24 May 2022 - 08:39 PM

This is a screenshot of a non-linear (pre stretch) image that I took last night, so pretty much unprocessed, so don't look at it yet.  lol  I've tried DBE to get rid of the gradient at the top, but it doesn't do anything.  Any suggestions?  Do I need to stretch it first?  Thanks.

 

Capture.JPG



#2 johnpane

johnpane

    Ranger 4

  • *****
  • Posts: 345
  • Joined: 13 Dec 2014
  • Loc: Wexford, PA, USA

Posted 24 May 2022 - 08:47 PM

Did you select "Subtract" in the "Target Image Correction" section of the DBE process window?



#3 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 24 May 2022 - 08:51 PM

yes



#4 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 24 May 2022 - 08:53 PM

I think DBE thinks it is a nebula or something.  I guess I'll just crop it down.


Edited by c131frdave, 24 May 2022 - 08:58 PM.


#5 jdupton

jdupton

    Skylab

  • *****
  • Posts: 4,328
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 24 May 2022 - 09:13 PM

c131frdave,

 

   That does not look like a normal gradient.

 

   Did you take Flat frames and use them in the calibration stage? If the Master Flat were to overcorrect and there is an OAG prism in the optical train, then this would be an expected result. Fixing this with DBE is going to be very tough. It would be much easier to fix if it is caused by an over-correcting Master Flat.

 

   Have you used the Blink process to examine all the files (Lights and all calibration frames) to see if there is a bright area at the top. I have seen something similar sometimes but it would be exceeding rare for it to show up so prominently.

 

 

John


  • zakry3323 likes this

#6 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 24 May 2022 - 09:21 PM

Yes, it has flat frames.  I'm pretty sure it is the OAG mirror- I re-adjusted it prior to shooting.  I was hoping I could get rid of the gradient in post.  I've cropped it out and it's fine, though M51 is tighter than I had hoped.  I wanted to get those tiny galaxies around it in frame, but the frame would look weird with the top cut off.  Maybe I'll try again.


Edited by c131frdave, 24 May 2022 - 09:22 PM.


#7 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 24 May 2022 - 09:23 PM

I could fix it in photoshop!  lol  maybe I'll do that once I get done in PI.



#8 jdupton

jdupton

    Skylab

  • *****
  • Posts: 4,328
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 24 May 2022 - 09:26 PM

c131frdave,

 

   I don't see any prominent dust motes in the image you posted. You might try just reshooting the Flats and Flat-Darks to see if the issue goes away. You should also double check your Darks to be sure they exactly match the Lights. It is very common find that the cause of bad Flat calibration is actually due to the Darks used to calibrate the Lights.

 

 

John


Edited by jdupton, 24 May 2022 - 09:27 PM.

  • c131frdave likes this

#9 KTAZ

KTAZ

    Mercury-Atlas

  • *****
  • Posts: 2,784
  • Joined: 09 Apr 2020
  • Loc: Scottsdale, AZ

Posted 24 May 2022 - 09:27 PM

How are you using DBE? There are many approaches, some with very few sample points, and some folks use the "scatter" approach and plaster them everywhere.

 

For heavy gradients caused by a prism or optical components (not natural gradients) you might try division instead of subtraction and use limited sample points placed in the areas that you consider to be "normal".


  • jdupton likes this

#10 zakry3323

zakry3323

    Mercury-Atlas

  • *****
  • Posts: 2,927
  • Joined: 11 Apr 2016
  • Loc: Fairfield, Westmoreland County, PA

Posted 24 May 2022 - 09:48 PM

The dark circular background around M51 is also a tell-tale sign that your flats aren't working correctly. 



#11 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 28,932
  • Joined: 27 Oct 2014

Posted 24 May 2022 - 10:27 PM

I think DBE thinks it is a nebula or something.  I guess I'll just crop it down.

No DBE doesn't think it's a nebula or something.  DBE is a typical PI thing.  It only works well when you understand HOW to use it. 

 

You whistle up DBE.  Click reset at the bottom, it covers the image with a cross.  Click generate, DBE works better when you understand how to set the different parameters for that.  That creates a grid of sampling points (where the samples are not stars).  You maybe add a few more sampling points manually.  Set correction to "subtract".  Then you drop the DBE on the image.  PI gives you a background image, what it's subtracting.  Stretch it, examine it, and you can see if you need to adjust the sampling points.

 

I am not kidding.  That's how you get DBE to work well.

 

You've just had a great learning experience.  What you really need to know.  PI doesn't magically process better.  It gives you a wide array of very adjustable tools so that YOU can process better.  IF you know which to use when.  And how.

 

I have hundreds of hours of study and experiment in it.  Two 400+ page books.  Chapters in other books.   A weeklong small (individual attention) class with a world renown PI expert, Vicent Peris.  Peris spent maybe half an hour on DBE, showing us how to develop sampling points, and what good and bad backgrounds looked like.  Then had us try it on our personal data, coming around to critique our work.

 

I don't recommend PI for the majority of beginners.  Learning to use it takes too much time, time you need for learning other things.

 

I recommend Astro Pixel Processor.  Much easier.  Has an _excellent_ gradient reduction tool that's much easier to use.  A numbered workflow that walks you throught the process.  As I said, with PI you need to know which tool to use when.

 

I had imaged for over a year when I tackled PI.  This book was absolutely key.  I had tried web tutorials and failed miserably.

 

https://www.amazon.c...y/dp/3319256807

 

It is a great learning tool.  Walks you through a workflow, with operations of increasing complexity, in a well organized fashion.  But, it's not comprehensive.  There's only so much you can do with PI in 400+ pages.  <smiling, but still not kidding>

 

If you just follow an Internet website, PI is no better than anything else.  To get its value, you need to adapt it to your data.  Like by adjusting the sampling parameters/sampling points in DBE.  They'll be different for every image.  One size definitely doesn't fit all.

 

Minor point.  You use any gradient reduction before stretching.  Stretch the data, that distorts it, and makes the calculations work poorly.


Edited by bobzeq25, 24 May 2022 - 10:41 PM.

  • zakry3323 likes this

#12 ColdyNitesNC

ColdyNitesNC

    Mariner 2

  • -----
  • Posts: 276
  • Joined: 24 Feb 2018
  • Loc: Nevada City, Ca

Posted 24 May 2022 - 10:41 PM

You could just try ABE, and see what effect that has.



#13 KungFood

KungFood

    Apollo

  • -----
  • Posts: 1,052
  • Joined: 18 Jul 2014
  • Loc: Dallas, Tx

Posted 25 May 2022 - 01:21 AM

If you want to upload your stack (FITS preferred) myself and others can give it a go.



#14 charles.tremblay.darveau

charles.tremblay.darveau

    Mariner 2

  • *****
  • Posts: 261
  • Joined: 16 Oct 2020

Posted 25 May 2022 - 01:30 AM

Try increasing tolerance and see if it works.

#15 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:11 AM

I had many little boxes.  I had the size adjusted to 15, tolerance to 2, and auto generated.  Then I manually added sample points to emphasize the transition and get a good sample of the entire frame.  I'm at work now, but I believe I had 65 points.  I used subtraction and division, neither worked.  Keep in mind, this is pre-stretch- what you see is from the nuclear button.   I loaded the stacked image, hit the nuclear button, used ABE to get rid of the green, re-nuke stretched, then tried DBE to get rid of the band on top.  I'll try again tonight when I get home.  Thanks.



#16 kathyastro

kathyastro

    Aurora

  • *****
  • Posts: 4,578
  • Joined: 23 Dec 2016
  • Loc: Nova Scotia

Posted 25 May 2022 - 07:19 AM

Did you crop off stacking artifacts at the edges?  If you didn't and DBE saw contrast at the edge as a result, it will produce an edge like yours.



#17 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:22 AM

Yes



#18 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:24 AM

]I would say I cropped about 20% off the edges.  The total white section at the top is about 20% of the full frame.  It's my OAG prism.  I had to clean my prism, and when I put it back in, I accidentally put it in too far.  It didn't show up on the short preview exposure, but the longer exposures showed the problem.  Below is the full FOV of my camera on this object with 1000mm FL:

 

[attachment=2058063:Capture.JPG

Attached Thumbnails

  • Capture.JPG

Edited by c131frdave, 25 May 2022 - 07:26 AM.


#19 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:36 PM

I'm home.  I tried again increasing tolerance.  I can't get it to work.  If I go to a tolerance of 8, it will extract the green bar at the top, but it will be super black.  As I slowly decrease tolerance, it suddenly switches to green at the top.  There's no happy in between.  Below are my settings.

 

Capture.JPG



#20 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:37 PM

Here is a typical result:

 

Capture.JPG



#21 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:39 PM

 This is as good as I can get it.

 

Capture.JPG



#22 KungFood

KungFood

    Apollo

  • -----
  • Posts: 1,052
  • Joined: 18 Jul 2014
  • Loc: Dallas, Tx

Posted 25 May 2022 - 07:46 PM

One thing shouts at at me, because it was a problem I had that took some sleuthing by the fine folks here to suss out. Your data is in too low a bit rate. That is whats causing the posterized color gradient. What software are you using to stack, and what is your output format? If you are stacking to Tiff, it's very likely that it's outputting a 16bit Tiff, which is too low. ABE/DBE is then trying to sample at 32bit, which causes big ol' problems.

 

This isn't necessarily adding to your larger gradient problems, but it could be. Worth fixing first anyway.


  • c131frdave likes this

#23 jdupton

jdupton

    Skylab

  • *****
  • Posts: 4,328
  • Joined: 21 Nov 2010
  • Loc: Central Texas, USA

Posted 25 May 2022 - 07:46 PM

c131frdave,

 

   Since you know that the "gradient" is due to an optical issue (bad Flat calibration), then you must use division in the correction. If you repeat your experiment but use division correction, you should have slightly better results.

 

   Hover the mouse over the Correction entry and read the tool tip. Subtraction usually works best on gradients that associated with the Lights while division is best used for multiplicative optical defects like vignetting (and OAG shadows). In this case, the abrupt nature of gradient probably also needs a smaller value for the Smoothing Factor in the Model Parameters section. It defaults to 0.250 but try several values in the range of 0.050 to 0.200 and see what happens.

 

   Still, DBE may not be up to the task. The best course to save your session of Lights may be to try regenerating the Flats to see if the gross overcorrection goes away.

 

 

John


Edited by jdupton, 25 May 2022 - 07:49 PM.

  • KTAZ likes this

#24 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 07:55 PM

Thanks guys.  It is 16 bit.  I used DSS because I can't get PI to stack the subs.  I tried what Adam Block calls the "Brute Force" method, and created 240 sample points and got the below.  I'll try to save my stack as a 32 bit file and see how that goes.  This isn't too bad, I can work with it, but if 32 bit works better, I'll take that route.  Thanks!!

 

Capture.JPG



#25 c131frdave

c131frdave

    Aurora

  • *****
  • topic starter
  • Posts: 4,696
  • Joined: 17 Jan 2005

Posted 25 May 2022 - 08:16 PM

Using the 32 bit source image is better.  But it's still there.  I think I can work with it though.  Thanks for the heads up on that.  I'd retake my flats, but I'm selling my camera and I have it all packed up and broken down, so there's no way to duplicate exactly how I had it configured.




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