Struggles with Stars in Startools

[robonrome]

Hi all, really hoping I can get some help with my Startools processing... I've simply never been able to get nice natural looking stars using the tool. How I process inevitably seems to end up with tight blown out cores and extended diffuse glow around these cores for moderate to bright stars... thus far I've used a bandaid approach of blending in stars from RGB combined TIFs out of Astro Pixel Processor (APP, which is what I use to produce the linear stacks for Startools as well) in PS as a final stage.

I'm sure it's user error on my part and I've seen plenty of Startools processed images with natural looking stars, so seeking help here.

To illustrate below is a 300% extract of around 6hr SHO data after I've run through my usual workflow in Startools (binned 50%) and the exact same area after basic RGB combine in APP alsobinned 50%. I like the nebulosity I get in Startools but here I would the blend these two in PS to create more natural looking stars.

I confess to a pretty lazy approach in Startools tending to use default settings in most cases.
1. compose SII red, HA green, OIII blue
2. autodev select linear data and use sample area when data is overstretched
3. bin usually 50% sometime 70% if more data
4. crop outer edge
5. wipe - narrowband preset
6. contrast - default
7. HDR depends on subject but usually reduce strength to 1.1... optimise soft for example here
8. sharpen - auto gen star mask... invert and grow any blobs that dont cover star glow and reinvert and apply default
9. deconv - retain same mask and usually default 1.5 or slightly higher up to 2.3 or so
10. colour - SHO presets... typically use detail aware
11. reduce stars - tighten but reduced iterations to 6 or so at most.
12. superstructure ~40%
13. noise reduction usually end up with around double what the default pixel setting comes up as.

Would greatly appreciate any help I could gain here to produce images from Startools with nice stars.

My Astrobin images can be found at link below... most are Startools processed but typically with some blending of APP RGBcombined TIF to try and improve stars.
https://www.astrobin.com/users/robonrome/

[Mike in Rancho]

Hi and welcome! That's a very nice set of bold images and great targets on your a-bin's main page there.

I'm pretty sure I've seen this topic come up before. I guess some of it depends on what you consider "natural" or "nice" stars. One could say that natural is pinpointed as much as reasonably possible down to the airy disk/pattern, surrounded by diffraction. Or it could be more of a soft blob. I think ST generally follows the former philosophy, based upon what you can achieve with the AutoDev and deconvolution.

I imagine there might be ways to apply more of a soft stretch to stars, so that they look more like "levels and curves" stars, though I haven't gone looking for that myself. Ivo will undoubtedly know.

If desired for aesthetic purposes, deconvolution can be sort of reversed using Layer and the undo buffer, perhaps with a little fuzz and kernel to ease the transition. Deringing can also be applied in deconvolution (I can't recall 1.7, but works great in 1.8), or independently in the Shrink module - with no iterations if you don't want to dim or tighten the stars.

One thing I notice from your general workflow, which may just have been skipped, is that I only see one AutoDev, and it is not post-Wipe. Generally the initial AutoDev is a quick look-see at your data so that you can make decisions like cropping or rotation, and it's also nice to see the target on screen when trying out the Bin levels. But it is not necessary, and if I am going through known data I'll just go straight to my crops and bins, and then into Wipe.

After Wipe is when you'll want to apply your "final" (you can always go back later) AutoDev, and is the one where you will want to adjust ignore fine detail and your region of interest, if applicable.

What I do, and I believe it may be a general rule of thumb, is to increase the IFD until the background stops getting darker. It doesn't always work, every stack is different, and sometimes I'll go beyond just a wee bit which usually brightens the main target up. After the IFD is set, I will then decide on and set the ROI.

Your two samples seem to have differing strengths of stretching applied, stronger on the ST image I would hazard, so it is a bit difficult to compare, other than as generally noted above.

Hope that's of some help? You could always link some stacks too so that they could be tried out and a sample workflow or log provided.

[hixx]

Hi robonrome,
there are a few important things to get a reasonable ST output:
- Wipe needs to be aggressive enough to remove the background: In Your image, a brownish haze is still left over and I suspect your background is not well calibrated by Wipe (background calibration should NOT be done in APP stacking!)
- Mike is right, there should be a second global stretch (Autodev) to leverage the dynamic range freed up by Wipe, followed by local dynamic range optimization (contrast, HDR).
- I strongly recommend ST V 1.8 - in regards to stars, I feel SVDecon provides greater results than previous Decon versions. Improvents also inlclude Wipe, HDR, Sharp and Contrast
. Shrink and SuperStructure modules may be mighty tools addressing stars, but it is easy to overdo things, ending up with a "processed" look
cheers,
jochen

[robonrome]

Mike thanks so much for this detailed and very helpful reply. I think I must most definitely be in the soft blob camp

Yes I missed listing the second autodev step (Mainly cuase I dont think you can really see anything or proceed without that step after WIPE anyway), but have learnt quite a bit from your post - one mistake I've been making from what you have shared is I am applying ignore fine detail and ROI on the first stretch and that might be behind why my wipe appears not to be doing its job as Jochen mentions.

So I'll try your approach and set on second stretch... I note you do IFD till no more darkening background (nice tip) then ROI... what exactly are you looking for in ROI application? While I love the structured and simple workflow of startools this is one of my struggles...not really knowing what I should be looking for at certain stages... same could be said for Wipe and Noise reduction I see stuff happening but have no idea if it's good bad or indifferent... more tips like your IFD till background stops darkening would be great.

And some great advice on how to address the stars, although I suspect I may need a little more hand holding. I will as you suggest get some stacks dropped on cloud and share when I get a moment.

Thnaks again,

rob

Hi and welcome! That's a very nice set of bold images and great targets on your a-bin's main page there.

I'm pretty sure I've seen this topic come up before. I guess some of it depends on what you consider "natural" or "nice" stars. One could say that natural is pinpointed as much as reasonably possible down to the airy disk/pattern, surrounded by diffraction. Or it could be more of a soft blob. I think ST generally follows the former philosophy, based upon what you can achieve with the AutoDev and deconvolution.

I imagine there might be ways to apply more of a soft stretch to stars, so that they look more like "levels and curves" stars, though I haven't gone looking for that myself. Ivo will undoubtedly know.

If desired for aesthetic purposes, deconvolution can be sort of reversed using Layer and the undo buffer, perhaps with a little fuzz and kernel to ease the transition. Deringing can also be applied in deconvolution (I can't recall 1.7, but works great in 1.8), or independently in the Shrink module - with no iterations if you don't want to dim or tighten the stars.

One thing I notice from your general workflow, which may just have been skipped, is that I only see one AutoDev, and it is not post-Wipe. Generally the initial AutoDev is a quick look-see at your data so that you can make decisions like cropping or rotation, and it's also nice to see the target on screen when trying out the Bin levels. But it is not necessary, and if I am going through known data I'll just go straight to my crops and bins, and then into Wipe.

After Wipe is when you'll want to apply your "final" (you can always go back later) AutoDev, and is the one where you will want to adjust ignore fine detail and your region of interest, if applicable.

What I do, and I believe it may be a general rule of thumb, is to increase the IFD until the background stops getting darker. It doesn't always work, every stack is different, and sometimes I'll go beyond just a wee bit which usually brightens the main target up. After the IFD is set, I will then decide on and set the ROI.

Your two samples seem to have differing strengths of stretching applied, stronger on the ST image I would hazard, so it is a bit difficult to compare, other than as generally noted above.

Hope that's of some help? You could always link some stacks too so that they could be tried out and a sample workflow or log provided.

[robonrome]

Thanks Jochen, to clairify I am using 1.8 (poorly no doubt and have struggled with SV decon) and do run a second stretch after wipe. I always turn off background neutralisation etc in APP and save and use only unstretched stacks. As I note above back to Mike though I've likley been mucking up the stretch process and in turn likely affecting WIPE as you mention...although i remain pretty clueless as to what I am looking for with WIPE, just something that doesnt have harsh transitions or vignetting but beyond that... when imaging a new region (as here) it's also tough to know what's due to poor wipe and what's nebulosity. I'd welcome any rules of thumb on various modules and settings in general and with respect to a result that ends in more natural stars (which to me means - no hard central to outer glow transition). Thanks again. rob

Hi robonrome,
there are a few important things to get a reasonable ST output:
- Wipe needs to be aggressive enough to remove the background: In Your image, a brownish haze is still left over and I suspect your background is not well calibrated by Wipe (background calibration should NOT be done in APP stacking!)
- Mike is right, there should be a second global stretch (Autodev) to leverage the dynamic range freed up by Wipe, followed by local dynamic range optimization (contrast, HDR).
- I strongly recommend ST V 1.8 - in regards to stars, I feel SVDecon provides greater results than previous Decon versions. Improvents also inlclude Wipe, HDR, Sharp and Contrast
. Shrink and SuperStructure modules may be mighty tools addressing stars, but it is easy to overdo things, ending up with a "processed" look
cheers,
jochen

[Mike in Rancho]

Hi Rob,

Yeah that's the thing about ST. The fairly friendly interface, available step-by-step workflow (as Ivo would say, reveal from large to fine scales in that order), and that beginners can often get a decent result just by walking through the defaults, tend to get it tagged as a simplistic processing software. But there's a ton of power under the hood, and along with that comes a decent learning curve. Which I am still very much on, by the way.

It does take time and practice, which you will build up by going through many different datasets. To help with that, in addition to your own, you might also try out user uploads and available top-notch data, all to get a feel for how things react and look.

On the blob stars, I am hesitant to recommend it, but you can always try FilmDev as your stretch after Wipe. This will not be an optimal stretch for dynamic range, however, like you can achieve with AutoDev. In fact, if you think about it, if your stars are blobs, your structural detail may very well end up that way also, and as to both stars and target, you are missing out on one of ST's advantages. A resolved star should be a point source, surrounded by diffraction. Or so I think I am coming to understand. Ivo would have more knowledgeable advice on this subject.

Wipe is just to set the field evenly and look for, plus fix if possible, flaws. So, as you say, vignetting (hopefully taken care of by flats). Gradients. Look for LP gradients and edges, and raise aggressiveness to try to even those out. For NB, you can start with the NB preset and go from there. Raise DAF if needed for cold pixels, or color blotches. It gets a lot more complex. Perusing the online module descriptions, user notes here on the forums, and pdf unofficial manual, can all be of help.

Wipe also handles color casts, and sets up your data to go into the color module later. As Jochen noted, dynamic range freed up can then be used in your stretch. As such, it is not optional and must be part of the workflow, though if you have no gradients or other such issues that you believe need addressing, you can pretty much zero out all the settings.

Don't use Wipe to attack noise. What you are looking for is an even field (though because Wipe hyperstretches the data, it will often "look" quite noisy - not to worry).

The AutoDev, with or without ROI (and other settings) takes practice to get a feel for. But you can always go back and amend your AutoDev stretch at any time while still in tracking. As with any AP processing, too much global stretch can raise noise that you may have difficulty stomping out later, even though there are a lot of opportunities to do so (Contrast module shadow linearity, and several of the SuperStructure presets to name a couple) in conjunction with ST's own noise tracking, which get implemented at the very end.

Well that's probably enough to chew on for now, eh, and no need to get into SVD? A lot of the rest you've probably got a bit more of a handle on anyway based on your use history (those as I am learning, all the modules are more complex than one ever thinks in the beginning). And it's not uncommon for Wipe and AutoDev to remain somewhat perplexing even with a moderate bit of ST time under the belt. For sure was that way with me, before I started getting more comfortable with them.

[hixx]

Hi robonrome, I think not using IFD in the first stretch would make no difference. The important thing is to REDO the stretch when selecting second time and use IFD then. In first strectch all You need is to see where the noise is -you need to do nothing here.
in Wipe, use the Correlation filtering to increase SNR, mask out dark anomalies and use enough aggressiveness - try presets if unsure. And Mike is right, don't use Wipe to attack noise - Wipe is to attack gradients, amp-glow, LP etc with low to moderate undulation. In fact if done correctly, Wipe (plus the temp Autodev will reveal the noise floor (usually the shot noise). If you think about it, this makes sense because now you have revealed all detail captured from the noise floor up to white in one global stretch you can now refine locally using Contrast, HDR, Sharp & SVDecon. The shot noise may be tackeled by Denoise, Super Structure, Flux later on (I use most module just a tad to avoid visibility of the denoising "trick")
For SVDecon, check Guy's User Notes: I use to choose 3 stars in each corner and the image center and a few more halfways and along the image borders. Once selected, I switch resampling mode to centroid tracking. This usually creates snappy stars. To accelerate the learning curve a bit, You may have a look in the Tutorial section of the inofficial manual. It contains step-by-step instructions covering modules for the beginner's workflow followed by advanced workflow up to special techniques (for Mike et. al. ). I recommend using the (flawless) IKI datasets from homepage to train these techniques without the hazzle of capturing or stacking imperfection
good luck,
jochen

[Mike in Rancho]

Hi robonrome, I think not using IFD in the first stretch would make no difference. The important thing is to REDO the stretch when selecting second time and use IFD then. In first strectch all You need is to see where the noise is -you need to do nothing here.
in Wipe, use the Correlation filtering to increase SNR, mask out dark anomalies and use enough aggressiveness - try presets if unsure. And Mike is right, don't use Wipe to attack noise - Wipe is to attack gradients, amp-glow, LP etc with low to moderate undulation. In fact if done correctly, Wipe (plus the temp Autodev will reveal the noise floor (usually the shot noise). If you think about it, this makes sense because now you have revealed all detail captured from the noise floor up to white in one global stretch you can now refine locally using Contrast, HDR, Sharp & SVDecon. The shot noise may be tackeled by Denoise, Super Structure, Flux later on (I use most module just a tad to avoid visibility of the denoising "trick")
For SVDecon, check Guy's User Notes: I use to choose 3 stars in each corner and the image center and a few more halfways and along the image borders. Once selected, I switch resampling mode to centroid tracking. This usually creates snappy stars. To accelerate the learning curve a bit, You may have a look in the Tutorial section of the inofficial manual. It contains step-by-step instructions covering modules for the beginner's workflow followed by advanced workflow up to special techniques (for Mike et. al. ). I recommend using the (flawless) IKI datasets from homepage to train these techniques without the hazzle of capturing or stacking imperfection
good luck,
jochen

Yes, I bypassed Advanced and went straight to Special. Primarily because I'm a goofball and can't resist fiddling with things I don't yet understand.

Good post though, Jochen, I am learning some new stuff myself.

I did forget about that from Rob's post. Indeed the initial Autodev should not be not changing anything permanently, so Wipe will do its Wipe thing no matter what you did in A/D before that. It's all about the time-shifting.

I have not yet employed Flux, so that will be a new ST module for me to figure out. But, what is the avoid visibility of the denoising trick? This sounds like something I must learn!

Your SVD steps are reasonably close to my own method too, so good to know I've been on the right track.

Oh and I just downloaded the unofficial manual again, as I realized I only had it on my laptop. Will have to check it out for new tricks since the 1.7 manual.

But while on the hunt for it around the ST website, I did run across a strange Commodore 64 thing I couldn't quite figure out.

Oh well.

[hixx]

Hi Mike,
Ok - off topic: there are multiple ways to fight / mitigate noise:
1) Correlation Filtering in Wipe.
2) Good global stretch with reasonable ROI and IFD (can be redone later on the processing)
3) I strongly reduce Shadow Dynamic Range Allocation in Contrast. This pulls noise in dark areas towards black.
4) For shot noise or chroma noise /hot pixels , elpajare described a good method using Flux (only required with excessive noise - now also in user notes and special techniques
5) Isolate or Dim Small presets in Super Structure module

6) only the remainder will now need Denoise - every step is just removing a certain type of noise impacting signal just slightly, so its processing does not stand out while the overall improvement will become visible
7) reduced Dark Saturation in Color helps keeping Chroma noise down
8) Most important: more Integration time will fight shot noise best, Dithering fights walking noise and also a high number of Darks, Flats and Darkflats will help, because the noise of Calibration files will effectively be added during Stacking

Cheers,
Jochen

[Mike in Rancho]

Thanks, Jochen

Yes, I am the king of going off-topic. I can't help myself. My apologies to Rob.

Thank you for your list, it is a nice round-up. I already know of and frequently do (depending on what's in front of me) all 8 of those, except for Flux. I will look for the Flux method in the techniques sections in case that ever needs added to my arsenal.



PS - hopefully your list is also helpful to Rob, as it can be integral to really helping out an image, especially if one were still struggling a little with Wipe and A/D. I will add that in Contrast, in addition to lowering the shadow dynamic range, simultaneously lowering the locality control can help "return" faint nebulous regions of main target - in essence backing off application of heavy contrast therein. YMMV and depends on the dataset.