StarTools

Hi everyone
Canon 700d with sensor at 36º -which may be part of the issue-.

Here's a post I sent to the sgl forum

Hi everyone. I'm trying to diagnose the irregular darkish patches appearing on this snap. I think it may be produced by one of the st denoise setting -I'll ask specific questions on their forum- but am posting here just in case anyone has seen the effect before. The only quirk is a, lp filter but I've eliminated that as the cause. I've hit the contrast to bring out the dark areas. Some say they can't see what I'm on about; you can see the effect better by standing back from the screen. Maybe I'm imagining it... TIA for any pointers and clear skies.

If anyone can see what I'm on about (many can't, sorry) It seems to happen just after denoise; a lower smoothness reduces the effect but it takes ages on my computer to trial and error it. Ironic too as a friend has lent me his telescope whilst he's away. Yeah, I know. Too many variables. Just wondering whether anyone can help.
Cheers and clear skies,
Steve
** Just checked: the patches kick in after the denoise step
***fits: https://drive.google.com/open?id=0B_uUt ... HhDX0t6aGs

Hi Steve,

What version of StarTools is this on?

I'm having some trouble replicating this. Could you post a before/after denoise example at full resolution? (a crop will do)

Thanks!

Hi Ivo
st 1.4.315
https://stargazerslounge.com/topic/2769 ... k-patches/

Here are some before and after denoise on full sized tif:
https://drive.google.com/open?id=0B_uUt ... UJZbU9TaEU
The filenames are self explanatory. After the decon, you can see where the dark patches will form.

TIA and clear skies,
Steve

Here they are in another post:
http://forum.startools.org/viewtopic.php?f=10&t=1037
Cheers,
Steve

Hi Steve,

I've had a chance to look at your TIFFs now. The darker patterns are definitely a feature of your data. Sometimes they can be detail (dust lanes, variations in stellar density), but they can also be artefacts of some sort (under or over correction by your flat frames, some type of noise reduction, failure to dither properly).

For what it's worth, without having seen the progression and the "appearance" of the structures in the various TIFFs I was successful in seeing them from the start (which I consciously tried to do).

True Gaussian (and - by approximation - the Poisson noise we mostly deal with in AP) exhibits random noise at all scales, its amplitude becoming smaller and smaller (and thus harder to detect) as scale increases. You can see this effect in the Denoise module's first step - which exploits exactly this - where StarTools asks you to specify the size of noise grain beyond which you can't see its existence anymore; StarTools is filtering out bigger and bigger scales as you increase the noise grain parameter.

Given the Poisson nature of the noise we tend to deal with, noise reduction in StarTools indeed deals with noise grain at different scales (and cleverly exploiting psycho-visual effects that go along with that). Correlating detail at bigger scale tends to show on a smaller scale, but much less often the other way around. In your images, we see correlation at larger scales that doesn't appear to be present at smaller scales; somewhat suspicious.

Looking at the Veil data again, there is definitely "clumping" (e.g. correlation at large scales"). There's actually a filter in the Layer module that seeks out this "clumping" across the 3 channels called ("Local Maximum Entropy RGB Selection"). Turn it on and you'll see something like this;
The clumping appears to be somewhat elongated (diagonally), which leads me to believe we're looking at a dithering problem; the dithering displacement may not have been sufficient or the frames that were selected during stacking were to heavily weighted towards one particular amount of displacement.

I'd probably suggest using an amount of dithering that is much larger than what you're currently using, while making sure that if you stack frames a good distribution of frames are selected for stacking across the full displacement range (e.g. if you're spiralling out, make sure you don't just stack the frames from one part of the spiral because the seeing happened to be best then).

Let us know how you go!

Hi. Here we are with random dither of 20 as opposed to 10.
http://4.bp.blogspot.com/-_EdiOy3U5hM/V ... 0/b168.jpg
Definite improvement.

I think the lack of randomness you mention maybe due to backlash. If the dither request hits dec by only a small amount, the backlash isn't cancelled and the correct distance isn't applied. I'm turning grub screws almost as I speak and the PHD guys have suggested I enable backlash compensation.

BTW, really pleased with the snap. I'd been trying for a dark nebula for ages. The dither linked to the ongoing fringe killer methods: http://forum.startools.org/viewtopic.php?f=6&t=924 has paid off. Without StarTools I'd be nowhere.
Thanks for your time,
Clear skies,
Steve