ABE on Individual Sub Before Stacking?

[Bmegdal]

I am fairly new to PI, so this may have been discussed before or be just plain wrong.  It seems to me that because gradients due to light pollution vary as sub frames are taken at different positions in the sky, it would make sense to run ABE (or DBE) on each subframe before they are combined into a master of each color.  Once the sub frames are combined (e.g. all Red subs stacked), the composite background gradient is quite complex and therefore harder to remove, while in the individual subs the gradient is much simpler.

I have found this easy to do with ABE and an image containers, but wonder what others think of this idea?

Thanks

- Barry

[pfile]

yes - have definitely done this for images with bad LP. i think some images i've made would be impossible without employing this technique.

supposedly DBE is getting a rewrite that would allow it to be used with ImageContainer.

rob

[Bmegdal]

Thanks -- good to hear that the approach makes sense.  It is pretty quick to do, do not sure why it shouldn't be routine.

- Barry

[dnault42]

Hey Barry,

Some tips on doing this:
- I've tried this before and after calibration and had much better success doing it before calibration
- Do a maximum integration of all your un-registered images.  I started doing this specifically for DBE so that I could find sample locations that didn't overlap any stars.
- Test your ABE parameters with that image
- I would use a very low function order so that it only catches gradients and not things like vignetting (which can persist after calibration in the presence of strong LP gradients).  Something on the order of 1 or 2.
- Make sure to set it to subtraction mode and enable the normalize option so that the background levels are maintained

After you do your pass of ABE then run your normal calibration, registration, integration, etc.  I've had good success doing it this way, but certainly trying other options is good and may provide you with better results for your data set.

Regards,
David

[Bmegdal]

Thank you for the suggestions--  will give them a try.

- Barry

[jerryyyyy]

My two cents as this is a good question.  I used to do this but then decided to do DBE on RGB together, just saves a few steps. 

But what i do first is to do DBE on the L image and to make sure none of the samples overlap a star.  You can usually see this obviously. 

I use a 12x12 or 10x10 matrix but only two rows and columns on the edges.... using the L created DBE matrix

Thus if you have funny gradients that differ by filter this patter (IMHO) will cover them all... not sure of the math but if a regression is calculated using those points across the image it should correct most gradients that go across the image gradually, which is what I get in my LPed area....