Uneven background

Maxim:
Looks more like an artefact of heavy noise reduction, so called "orange peel" effect. Did you use noiseXterminator? To be honest I don't know a definite way to prevent it. You might try diffrent noise reduction techniques, e.g. TGVDenoise process.

Thanks Maxim. Yes, I ran through the usual PI workflow after ImageIntegration: DBE, BX, NX, and HT (carefully). I thought about possible over-stretching, but I don't think that's the case - I ran it through twice. I'll give TGV a try.

Heavy noise reduction and not perfectly flat background. TGVDenoise won't help you here. Take a step back and see what you can do to avoid a bright backdrop.

Try HT *before* NX

Hi all,
Well, your suggestions certainly improved things about 93.665 percent - many thanks. After some experimenting with all your ideas, this was my final attempt:
After running the usual Integration and DBE, I applied BX, Correct only (as usual), then BX normal at 70%. Then applied NX at 30%, ran the result through HT carefully, then applied NX again at 50%. This is the result - certainly googols better than before. A little further delving, and I'll be good. Lesson learned: do not be misled by default settings.

Thank you all, again for the wonderful thinks.
 - - Steve

@Steve Solon and Terry Chatterton BTW, how long were your exposures? 5 minutes? 10 minutes?

Victor Greene:
Id like to add that it appears your image has not been cropped. The entire right edge of the image is a result of the stacking process and is visibly darker than other parts of image. I’d crop in a bit and give it another go. Uneven backgrounds will certainly lead to problems further down the line.

Hi Victor. Interesting idea. I usually save cropping for the very last process. Are you suggesting that the overlapped areas would pollute the HistogramTransformation readings? Hmm - never thought of that. Ah, always learning. Thank you for the idea. I'll give it a shot.
- - Steve

Best of all is to use LocalNormalization to produce normalization files and add them to the stacking images. If the frames are properly scattered around the final stacked frames shouldn't have any dark border at all.

Jon Rista:
I also wonder about the periphery artifacts. It doesn't look like you cropped out the stacking artifacts (areas of the image border that have varying SNR due to shifts in the frames, either due to dithering or drift). I would crop out all that edge junk first, then I would try to reduce noise without obliterating it, and see if that not only helps with the mottling, but also helps preserve fine details.

Thanks, Jon. As I mentioned to Victor, I usually do crop as the final step, but it appears that the uneven border edges of an uncropped image can have an overall effect, so I will crop first. Also, LocalNormalization, as Andrea mentioned, is a step I've not used before (one of PI's thousands), so I will give it a go, as well. Thanks again.
- - Steve

Steve Solon:

Jon Rista:
I also wonder about the periphery artifacts. It doesn't look like you cropped out the stacking artifacts (areas of the image border that have varying SNR due to shifts in the frames, either due to dithering or drift). I would crop out all that edge junk first, then I would try to reduce noise without obliterating it, and see if that not only helps with the mottling, but also helps preserve fine details.

Thanks, Jon. As I mentioned to Victor, I usually do crop as the final step, but it appears that the uneven border edges of an uncropped image can have an overall effect, so I will crop first. Also, LocalNormalization, as Andrea mentioned, is a step I've not used before (one of PI's thousands), so I will give it a go, as well. Thanks again.
- - Steve

The janky regions around the edges can definitely cause problems with various processes. Its best to crop them out as early as possible. I always do it right after integration. That'll improve things like STF even, as well as anything that evaluates things at various scales, etc. ABE can be affected by it, DBE could be if you placed any samples in that area, etc. etc.

Regarding LocalNormalization. That is a process that Juan mentions is a more complex process that should only be applied if it is truly needed. I used to know how to use it, but it was changed rather dramatically in newer versions of PI, and I am no longer sure how to produce useful results from it. I do often have thin cloudcover, and that is the kind of thing that LN is supposed to be able to help normalize out. In the past, it took a fair bit of trial and error. With the new tool, I've not been able to figure out how it works, and every time I've tried to use it, regardless of any adjustments I've made to its settings, I never see any difference with vs. without it after ImageIntegration. So, its not entirely a necessary step, and its primarily intended for when you have inconsistent backgrounds in your frames. It usually requires having at least one frame that is very good, good enough to be the reference frame, so that the variations in the background levels can be identified.

In any case, I think your primary issue is overly aggressive NR. Even with modern tools, that destroys smaller scale nuanced variations in signal level. The mottling is often an intrinsic characteristic of the stack, even...its not necessarily that it was introduced by NR, but more that it was revealed by NR. Without sufficiently aggressive dithering, integrating signals with a poisson/gaussian noise distribution, will often have a bit of that mottling inherent to the signal in the final integration. Dithering needs to be sufficiently aggressive (needs to move the frame enough) each time in order to produce an even noise distribution in the integration. This mottling is also the result of integrating, you won't see it in the individual subs, its an artifact that results from the nature of weakly dithered noise with a poisson or gaussian distribution. I always start with dithering enough that each light frame moves about 10 pixels or more, but you may need even more dithering than that to actually eliminate the mottling. 

If the mottling is indeed an intrinsic characteristic of the signal in your integrated light master, all that your NR is doing, is revealing it. Its not really adding the mottling. Some forms of NR can also add mottling, especially multi-scale forms of NR. At medium scales, some NR routines can introduce medium-scale artifacts, in which case then the NR IS in fact ADDING the mottling. A multi-scale NR tool can be a means of reducing the mottling. Since it usually is a medium-largish scale artifact, you could use something like MMT to target that/those scales, and try to even out the mottling effects. At the same time, reducing but not obliterating noise at finer scales, can help preserve smaller scale structures and help hide the mottling.

One thing I leaned about NoiseXterminator is that you can use it on linear images but the neural net was trained for non-linear images.

This went against my normal process flow of doing noise reduction before stretching but once I moved NoiseXterminator to non-lenear processing the blob/orange peel background went away.