DBE pitfalls

[sixburg]

Does DBE remove valuable data?  What I've been doing lately with DBE is to set a rather large default sample radius (of 30 or more), 20-25 samples per row and let DBE generate the samples.  After some very minor removal of samples on nebulosity I just go with it.  I've tried to place samples on just the background or as some tutorials say in the corners (I suppose they have vignetting or something), but manual placements usually don't work well for me.

Does my approach remove wanted data?  If so, I can't really see it.  I've done noise evaluations before an after as a check to see if anything substantive changed in the image although I suspect this isn't the best way to see if data has been lost. 

Isn't DBE in part dependent on image scale?  I mean with other sensors the settings could be completely different, right?  I've noticed that my approach with my set up and location results in very similar settings.  Same goes for TGV and other tools as well. 

[jkmorse]

Sixburg,

What you need to watch out for is placing any boxes on signal you want to keep, especially if it is a weak signal such as faint nebulosity or faint galaxy arm features.  If in doubt, remove any boxes so that the process does not treat the signal as something to be corrected.  In other posts Juan and other pros have suggested that you minimize boxes by just placing a few.  Another good guide is Harry's tutorials that have one specifically on using the DBE and is the process I follow most of the time. 

As to seeing whether you are removing data, that will not become evident until much later in the process workflow when you go to bring out all the faint stuff so just flipping between two DBE products may not show what you have lost. 

When I use DBE on binned image stacks I usually set the "generate matrix" settings at around 12/20/0.375 and then delete any that threaten areas of signal.  I usually up the first setting to around 20 for unbinned imager stacks.  Note, if you haven't discovered it yet, its easy to delete a whole set of boxes in a column by clicking on the top one you want to delete and just repeatedly hitting delete until you are done with that column (probably obvious, but it took me a while to find that shordut so I'm passing it along).

Hope that helps,

Jim 

[MortenBalling]

Especially when you use automatic sample generation, some boxes will cover a star. That gives a wrong background value in that box.

I normally click my way through every sample and check the preview, for any stars, showing up as dark spots. In that case I move the box slightly to an area without stars.

Also I recommend, that you look at a googled image of the object you're working with. M27 is an example. The faint outer nebulosity is easy to remove using DBE, without noticing. Having sample boxes to close to a galaxy, will kill the outer details of the arms. The sample boxes closest to the galaxy on either side, will be averaged to a value used for background correction. A better solution is to have a rather large distance from sample boxes to the object in the image.

Apart from that, DBE is awesome

Morten

[MikeOates]

It helps if you do a strong screen stretch that will then highlight where the faint nebulosity is, and then place the sample points avoiding those areas.

Mike

[EorEquis]

Of late I've become a  strong believer of quality over quantity when it comes to DBE.

I've found that often just 8-10 samples are enough..maybe 2-3 per quadrant.  I've further found that the conventional wisdom that samples must be placed on the edges and in the corners is often inaccurate...it's really only beneficial, imo, if the gradient in those areas is significantly stronger than that which is already covered by a nearby sample.

I use very large...20-30 pixel if I can...samples, and am very concerned with avoiding any nebulosity AND any star...not just bright ones, but ANY star.  As Mike suggested above...an "excessive" STF serves to bring up very faint nebulosity and details, which can then be avoided.

While that frequently leaves me with precious few areas I can place a sample, as mentioned above, it seems to produce the most desirable results, ime.

[sixburg]

Thanks for all the valuable insights.  I've read and followed so many DBE-related tutorials and while I won't go so far as to say they are conflicting I can definitely say that they are all different.  They range from using giant, overlapping samples, to placing just a handful, to clustering in the corners, etc etc. 

I experimented with the advice from EorEquis and have found the results to be better than my own approach which relied heavily on auto sample generation (on the order of 25 sample size and 25 per row).  The difference I believe is that all the samples are truly on background with either ZERO stars or very, very tiny ones...

Thanks,
Lloyd

[pfile]

the "edges and corners" thing is good for images of galaxies that don't have any IFN or other signal in the periphery of the image.

for nebulae, yeah, you need to seek out areas where there is no or very little signal and place your samples there.

the idea is that a gradient is going to be mostly smooth and go from one edge of the frame to another. hence the idea to sample at the edges. of course in reality things can be significantly more complex. if you shot something low on the horizon toward the light dome of a city, the gradient will change from frame to frame across the night, and your integrated frame will probably have some kind of U-shaped gradient in it. in that case it's important to get enough samples to capture that shape, but not necessarily all over the frame.

the most important thing (which RBA points out in his talks) is to carefully study the extracted background that DBE gives you and try to decide if it looks like the gradient you expect, or at all like a gradient. if it's lumpy there's a good chance that you've oversampled the image, or placed a sample on some bright feature which was not entirely rejected by the sample parameters.

rob