DBE against very strong gradients

[pfile]

i image from a very light polluted area, and if i shoot something near the celestial equator (like anywhere around orion) i end up with very strong LP gradients in my stacks.

DBE usually works wonderfully for me, but sometimes the LP is so strong that maybe half of the image will be "bad" samples. if i set the sample to "fixed" this allows me to proceed, but i can't change the weights from 1.0 in each channel. is this a bug? the results are OK with these fixed samples but looking at the final result i can tell there are still gradients around, which makes me wonder if i should try to change the sample weights.

or has this got nothing to do with it? what's the right way to use DBE in the presense of very strong gradients?

[Nocturnal]

Hi,

you can increase the tolerance number a bit higher. I sometimes go as high as 1.75. I'm not sure what gets screwed up when the tolerance is too high or if we should aim to keep it as low as possible.

[Harry page]

hI

if you look at my DBE video that I have redone I cover this problem 

Regards Harry

[pfile]

thanks guys, i'll try the tolerance and check the video.

[Juan Conejero]

Hi Robert,

if i set the sample to "fixed" this allows me to proceed, but i can't change the weights from 1.0 in each channel. is this a bug?

Not a bug. The statistical weights of a fixed sample are set to one per definition. A fixed sample is a sort of 'control point' that you define arbitrarily; there's no statistical basis to define the relative weight of an arbitrary data point.

Anyway you shouldn't need fixed samples. The advice on increasing the tolerance parameter is correct. An increased tolerance is necessary in cases where the dispersion of values is high, as happens with wild light pollution gradients.

[pfile]

thanks juan. and thanks sander and harry - i did watch the video.

i guess i was looking for a per-sample parameter to set rather than a global parameter. i don't know if it makes sense from the algorithm's point of view to set the tolerance on a per-sample basis, though. in the end i had to set the tolerance to 4.560 to pick up all my samples. the thing that worries me about this is that in the darker parts of the original image, the high tolerance would seem to be accepting stars as part of the background. i did my best to move the samples off of stars but there are too many stars...

i think this is a very extreme case. it was kind of an experiment - 8 minute subexposures of the orion area with a 50mm lens, through a CLS filter. the light pollution to my south includes all of silicon valley, so the gradient is huge. i managed to remove the gradient, but the final image has some color casts that i had to remove with SCNR. the extracted background looked reasonably smooth though.

[Harry page]

Hi

Of course there is the option of running dbe twice , the first to remove most of the error and the second run with

a tighter tollerance

Also a higher shadows relaxation number can help 


regards Harry

[pfile]

yeah, i have done double DBE... i guess the only problem (for me anyway) is that i start wondering if i'm subtracting signal. the 2nd pass at DBE seems to give something that is no longer a smooth gradient, and so it's hard to qualitatively evaluate the result.