Cannot remove all gradient using PixInsight DBE

A few comments :

- it seems you have way too much samples : don’t forget to try to keep your gradient « simple », it shouldn’t be too convoluted. I usually use between 5 and 8 samples per row. And therefore you should also keep the « continuity order » parameter of the model as low as possible, like 2. Except if you image with lots of streetlamps right next to you and have the moon right above you, and you image for very long time with these sources rotating as your mount turns… otherwise, the gradient shouldn’t that complex…

- all your samples are on the periphery of the image. Why ? You should avoid putting some on the nebula itself, but I see rooms where you can add some samples toward the middle of the frame as well. If they are « red » (ie. Pixinsight don’t take them into account to build the gradient that is subtracted), then increase again the tolerance until they are taken into account.

- you should remove the gradient on each channel separately if you use a monochrome camera.

Hope this helps !

Hi Andriy,

I have good experiences with applying ABE. I do apply it twice, first with division and secondly with subtraction.
I suggest you give it a try.

Cheers
Gernot

Your DBE has too few sampling points. I normally end up with a much higher density and then I remove them when they overlap with object. Youmight have t repeat the process several times to get rid of severe gradients.

Many thanks everyone for the recommendations, very much appreciated.
Will try every suggestion.
Thanks again!
Andriy.

Andriy,

I think that there is something disturbing in the imaging train. Could it be that light is comming in somewhere? Or maybe dew. Did you change the filter position due to flats? What about the darks? Do you use a dew cap? I had the same issue, it was light coming through a gap in the filter wheel. By the way, if raw data is so much bad like you showed here, you can not expect that DBE can solve this problem.

BR, Christoph

Jérémie:
A few comments :

- it seems you have way too much samples : don’t forget to try to keep your gradient « simple », it shouldn’t be too convoluted. I usually use between 5 and 8 samples per row. And therefore you should also keep the « continuity order » parameter of the model as low as possible, like 2. Except if you image with lots of streetlamps right next to you and have the moon right above you, and you image for very long time with these sources rotating as your mount turns... otherwise, the gradient shouldn’t that complex...

- all your samples are on the periphery of the image. Why ? You should avoid putting some on the nebula itself, but I see rooms where you can add some samples toward the middle of the frame as well. If they are « red » (ie. Pixinsight don’t take them into account to build the gradient that is subtracted), then increase again the tolerance until they are taken into account.

- you should remove the gradient on each channel separately if you use a monochrome camera.

Hope this helps !

Many thanks, @Jérémie. Have tried your advise and it seems to have helped a bit but the issue remains. Here is what I've got so far:

Also tried to put slightly more and slightly less samples across the field (except obvious nebulae areas) with no noticeable improvement. Will probably give it one more try and abandon. It was a great DBE exercise though, many thanks for your help.

Most likely the root cause had been in the imaging process, not sure yet what it was, I suspect it was due to either my clip-in light pollution filter or probably it was not properly fixed to the camera. Will do some experiments with it too. There was nothing like streetlamps or the Moon nearby.

Best regards,
Andriy.

Hi Andriy,

I can suggest you two additional procedures to deal with this kind of gradients.

1. Use DBE with fixed samples.

You can fix the values of the samples in DBE and this will give full control over the correction. The procedure involves to place a matrix of samples (as suggested, not too many due to two reasons, one, to produce a smooth mode, and two, because is difficult and tedious to tweak every value of every sample) and fix their values to a constant number, I use to use 0.50000. Then choose subtraction as a correction method and make an iterative procedure in which you change the value of one sample, check the result, and refine, until you achive the desired result.

2. Use Multiscale Gradient Correction from PTeam

This procedure has been developed by the PTeam and published in his website:

https://pixinsight.com/tutorials/multiscale-gradient-correction/index.html

The procedure needs the user to capture a wider FOV image in order to work. Maybe you have a wider capture from previous projects, but if not, you can use other sources from other users, the only restriction here would be that the wider FOV image shouldn't be too much processed. If you have any friend that has a wider image of this area and could share with you the master lights without postprocessing, would be ideal.

In my opinion, this second method is the best gradient reduction technique currently avaliable. It is awesome how well it works.

Hope it helps.

Gernot Schreider:
Hi Andriy,

I have good experiences with applying ABE. I do apply it twice, first with division and secondly with subtraction.
I suggest you give it a try.

Cheers
Gernot

Hi Gernot, many thanks, I've tried your advise too, did not help much, unfortunately.
Do you recommend applying division and then subtraction with the same Model Parameters and samples size/distribution or do you change them before subtraction? I've tried both options though.
Andriy.

Jérémie:
Hi @Andriy Kozhanov
Can you share your xisf file with us through Dropbox, WeTransfer or anything so that we can have a look ?
Can’t do it right now (going to work), but I can have a look this afternoon and make some tests in PI.
Have a nice day

Hi @Jérémie, yes, absolutely, here it is https://www.dropbox.com/s/02spmjlhudxjwae/IC405%20IC410%20integration-crop-colorslinearfit.xisf?dl=0
Really appreciate your help, but please don't spend much on this.
Andriy.

This is what I can achieve in about 15 minutes.

Left: your original

middle: after a pass of ABE with function degree = 1, and a slight crop.

right: after DBE.

You can take a look at the parameters I used. You can achieve better by carefully removing the sample points and by adding fixed sample points. (But you need to first understand what all these mean.)

As I said before, you really need to use very dense sampling to see any hope of removing the very irregular background.  The background in your image does not just contain a simple gradient.  There is something else that's quite wrong.  Something isn't right in either your imaging, or calibration and stacking.

Hi @Wei-Hao Wang, this is really great, many thanks for your help & advise! Will try it right now.

Yes, you are absolutely right, the root cause was in the imaging process, it had to do something with my clip-in light pollution filter (or it was not properly fixed) - will do some more experiments later.

Here is what I got with Optolong L-Extreme another day, the rest was exactly the same, except exposure 2 min with the LP filter vs. 5 min with L-Extreme:
- equipment
- pre-processing steps followed by colors linear fit (no ABE/DBE was applied to the image below):


This new image is plagued with other defects such as walking noise and a flats issue but that is a separate story. No such irregular gradient though.

Many thanks for your help again.

Andriy.