Lumpy background extraction

[TinySpeck]

This is a recurring problem for me.  No matter what I do with ABE or DBE I end up with a lumpy chrominance background like the attachment.  The lumps seem to be in the raw integrated data just prior to background extraction, too, to some extent.

I've tried many different function degrees in ABE and DBE, a range of sampling density and box size, various other settings, and multiple passes with either/both ABE/DBE.  The lumps seem to move around a little but they don't flatten out.  I can't seem to get rid of them in processing after background extraction either (i.e. BackgroundNeutralization or large-scale chrominance noise reduction).

Can anyone recommend any tricks for this?

[ngc1535]

This is a recurring problem for me.  No matter what I do with ABE or DBE I end up with a lumpy chrominance background like the attachment.  T
Can anyone recommend any tricks for this?

So out of curiosity... what tolerance did you use for DBE. With a high enough tolerance you will certainly make changes to the background (as well as your object).
Using DBE in this way is likely not the best method. Best on your picture- it appears the "lumpy" bit is the large scale mostly chromatic fluctuation (almost banding) that is
endemic to OSC/DLSR. Is this is the issue?

If you can distinguish the object from the background- there will exist a threshold below which no useful information is found- it is just the sky brightness itself. Below this threshold I suspect that simply desaturating the background will go a long way towards improving the result. This can be achieved with a good mask.

-Adam

[pfile]

one possible problem here is that DBE has a single tolerance value which is applied to all 3 channels of an RGB image. thus if you have one or two channels which have much higher background values than the other(s), you may end up increasing the tolerance so high in order to capture the weaker channel that the stronger channels are oversampled. the only way around this would be to do a pre-normalization of the 3 channels, for instance by breaking out the 3 channels and linear fitting 2 of them to the third, then putting the RGB image back together and then run DBE. or, another thing to try would be to first do a BackgroundNeutralization on the RGB image, then run DBE.

that might not be the problem though - as adam points out there can be other sources for these gradients which could be sensor-related. to his point about desaturating the background, one trick that i've seen posted here is to make a very strong mask which protects the stars and foreground object (with RangeMask) and then use the AtrousWavelets tool to remove the large-scale structures with the tool set to Chrominance mode. that will pretty much eliminate the large-scale color blotches in the image, but you have to be very careful as you can very easily desaturate the stars or your target.

rob

[TinySpeck]

Thanks for the help, Adam & Rob.  I am using a DSLR, and I didn't know that lumpy chromaticity was an endemic problem.  That could well be the root here, since I see it to some extent in my raw integrated image too.

I don't remember the specific tolerances I used for DBE, but varying that didn't help.

The suggestions about separating channels and linear fitting or BN before DBE are interesting and I'll give them a try.  I have tried masked aggressive MLT Starlet (same as ATrousWavelet?) on chromaticity without much success.  Desaturating with a mask is worth a try.  Good point about carefully protecting the real color though.  I love color in astrophotos and I don't want to wash that out.