DBE selected sample color map

[TinySpeck]

I haven't been able to find the meaning of the colors in the selected sample image in the top frame of the DynamicBackgroundExtraction process.  Are black pixels not considered in DBE's calculation of background level for that sample?  And what do the other colors mean?

[pfile]

black pixels are not considered. usually you don't want to see too many black pixels in that little window; it means the threshold is set too low for the samples in the vicinity of the selected sample to pick up any data. the sample will be red on the main image to show that it picked up no pixels.

the RGB view is a little confusing since you're looking at all 3 channels at once. so if for instance you see a pure red pixel, it means that no background has been picked up for the green and blue channels.

rob

[TinySpeck]

OK, thanks, Rob, that gives me a little more confidence in the auto-generated samples.  I've seen recommendations for making sure all of your samples have no stars whatsoever, but it sounds like DBE will suppress the fainter stars fine, depending on your settings.  I guess you would want to make sure the colored star halos are not included in your sample, though, so that they don't distort the background calculation.

[pfile]

faint, tiny stars are probably OK but for sure you gotta stay off bright stars and the haloes of bright stars.

rob

[TinySpeck]

Going through each auto-generated sample to make sure it doesn't contain a star or halo is really difficult when you have lots of samples.  It's made worse because of an ongoing bug in DBE where if you move a sample and hit the "next" key you're sent to some random sample, not the next one.  It seems like the way to avoid this is to adjust DBE settings so that your samples show black pixels for stars and haloes, if you can, so that's what I'm shooting for.  If I can do that along with placing a few samples manually where DBE has skipped placement it won't be too bad.

[pfile]

well for me, my images fall into two categories - narrowband, where the gradients are not terrible, and broadband, where the gradients are obscene and ridiculously complex due to multi-night projects. (but the target is usually a galaxy)

in the case of narrowband images (or RGB images of non-galaxy targets) i usually just place a few samples by hand. there's enough signal all over the image that any remaining gradients are not super visible (but they are definitely still there.) in this case there's no need to mass delete any samples as there are just a few.

in the case of RGB galaxy images, i just set the tolerance to something huge and let DBE auto place samples in a grid all over the image. if it misses the corners or something, i increase the tolerance and regenerate. then i just use my keyboard to delete all the samples over the galaxy. this is possible because the auto-generated samples are placed in columns starting at the top left of the image. so if you select a sample somewhere in the middle and start pressing delete, the next sample is always the one below and so pretty soon you can delete all the samples over the galaxy. then i remove the samples that are on the brightest stars and around the brightest stars.

at that point i'll turn the tolerance down and run "resize samples" (even though i havent changed the size.) this causes DBE to recompute whether or not the samples are bad. if the new tolerance is too low i increast the tolerance and repeat until i can capture all the samples again.

then l extract a background and do STF (maybe super-STF) and look for bright "lumps". these lumps indicate where samples are very likely on stars or star halos. i go back to the image under DBE and move or delete those samples and iterate.

so anyway the key to easy sample deletion is to let DBE generate them, but that only really works on galaxy-like images.

i agree that an astrometry-based sample placement would be awesome. PI definitely has all the internal hooks for this to be possible now.

rob

[TinySpeck]

My images are definitely in the obscene and ridiculously complex gradient camp, being a OSC galaxy guy in Bortle 7 Land at the moment.     Thanks for that DBE operational description, I'll try that.

[pfile]

yeah mono here but same problem, red zone. for me, AP would be impossible without PI's gradient removal tools.

i should mention that there are definitely people (maybe even juan included) that say that fewer samples are better for DBE. i actually agree with this, but i end up with LRGB images that have LP gradients and reflections and other terrible stuff that seems to be on the 300-500 pixel scale, which does necessitate a lot of samples to capture. there's always a possibility that you extract features from the background that are not actually LP. but for me, it's basically this or nothing so even if i end up oversubtracting features i have to live with that.

rob

[TinySpeck]

I hear ya, I have the same problem.  Sometimes when I compare my star colors to other images certain starts aren't right while most of the rest are.  I'm attributing that to a distorted DBE gradient, which I'm now trying to avoid.  I too need to use more samples than I'd like, to try and even out the wonky gradients I see.  I also progressively reduce my DBE Smoothing Factor to ~0.02, so the gradient which DBE comes up with ends up matching my samples very closely.