ATWT assistance in DBE

[j.w.]

Under the general heading of "everything old is new....", I've discovered what some undoubtedly have learned before me, that using ATWT to delete the smaller scales of an image is useful when using DBE.  Working on a recent image of M13 that had several difficult color/luminance gradients - I found that if I delete scales 1 to 32, DBE had a much easier time in picking those gradients out.  Using DBE, I produced only the background and subtracted that from a clone of the original image using PixelMath.  The results gave me an image background that I had worked much longer on to achieve without using ATWT first (however, this could be a user limitation - not just a new found efficiency! )

John

[RBA]

That's an interesting approach, thanks for sharing the idea!

Gradients are large scale structures almost by definition, so that makes sense.

OTOH, you'd definitely need to aim for very large structures, to avoid data "spreading" from mid-large scale object structures that might interfere with an accurate reading from the samples, and I think that could be the Achilles' heel in this approach.

I'll definitely play with that next time and share my experience, but let me take a rain check on that before I clear my concerns. I still think that overall, reading the actual data and setting the samples and parameters appropriately is the way to go, but I'm happy to change my mind if proven otherwise! In the end it may come to simply analyze the image, and see whether this method is suitable for that particular image or not, and M13 sounds like a possibly good candidate...

[georg.viehoever]

I am not sure, but maybe setting a higher "Default Sample Radius" in DBE would have a similar effect. Anyway, interesting idea.

Georg

[Nocturnal]

Hi,

it's my understanding that you can whatever processing you want to figure out the best places for your samples but you should not mess with the actual image you're applying DBE to. So if ATWT helps in sample placement you could create an image copy, apply ATWT and other processing, place samples and then move those samples to the original image. Perhaps I didn't understand the OP's proposal?

[RBA]

I tried it on some "old" data (from two weeks ago  ).
Both ways - using the ATWT'ed image just as a reference to use the "autogeneration" of samples, and actually building the background model out of the ATWT'ed image.

Neither approach worked for me. Large scale structures from good signal (objects) overwhelmed the field to the point the gradient was much less noticeable and not nearly as well-defined. Which is what I feared... Now, if this works for others, by all means, go for it.

Sander, I agree with what you say, but it's more fun to disagree, so I'm going to disagree as well.

The goal of DBE (for me at least) is to build a good background model that resembles the gradient in your image. If applying DBE on an "altered image" produces a more solid background model, so be it, right? I'd find that to be less "incorrect" - if done properly and in a way that makes sense - than applying DBE 3 or 4 times indiscriminately, a method often suggested here.

Of course all things need to come into place. You could alter the image in such way, that it wouldn't make any sense applying that background model to your original image.

Again I do agree with you in most of what you say, but "you should not mess with"... What? I LOVE messing up with... things!

[sleshin]

I gave this approach a try on the image I'm currently working on. It's a galaxy so the background is not complex ie there is no IFN etc in the background. The background model derived on the intact RGB master and the model from the image with the first 6 layers removed looked roughly the same. So in this one try, there seemed no advantage to removing small scale structures to help define the background model.

Steve

[j.w.]

I had actually tried increasing the radius - but was running into a lot of stars in autogeneration that required me to move a lot of things around.  In fact, that's where I hit upon the idea (blatantly stolen from various PixInsight tutorials, although mostly for mask creation).  I reasoned (not always a good thing) that since gradients are, in fact, large structures, removing the small structures that impact the measuring and re-creation of those gradients shouldn't have an adverse impact.  That said, due to the diversity and flow of the gradients, I wasn't trying to follow each gradient seam throughout the image - but to test what was possible using the automagical settings (and I wasn't dealing with any nebulosity, which would have complicated everything!).  I do agree, you are measuring something different.  I would suggest, though, that the gradient color and luminosity shouldn't be greatly changing since I am not severely altering either in creating the background model, just "blending" the two with the individual stars given much less emphasis (where points are not moved).  I did try using the points generated from measuring the ATWT modified image on the original image and came back to square one, with the points landing on a lot of stars and having to be moved.  In fact, by subtracting the ATWT produced background correction, it resulted in smoother transitions, especially around the brighter stars.  As I have said though, these results could very well be due to my lack of skill with DBE.  After producing various backgrounds for subtraction, and doing auto-STF (I do love "auto"!) on both the backgrounds and the resultant subtracted images - the ATWT produced background, and it's corresponding corrected image, proved better than any of my other attempts.

John