ST1.7 Decon and ST1.8 SV Decon

[admin]

Hi Martin,

I think the problem is (hopefully) clear now;

problem.jpg (63.67 KiB) Viewed 3152 times

My question is why my Apod Mask changes gets worse with 50% and 71% Binning ( I thought it would be the reverse and 35% Bin would not give a good Apod Mask )

Just like noise, any aberrant core pixels also get "cleaned up" (averaged out) at higher binning levels.

Different factors play a role with the correct detection of bright stars. The algorithm is certainly not infallible (that's why there are a lot of parameters to tweak in the Auto Mask generator, should it fail for a particular case). One factor at play may be that stars that are physically embedded in nebulosity (as I believe is the case here) may be harder to detect correctly if their light is diffused by the nebulosity in front of them, making them look like "fuzzier" than stars directly in front of any nebulosity.

Regardless, none of this should be a major problem as you can touch up the mask with - literally - a single click in the Mask editor for each star;

StarTools_2840.jpg (95.52 KiB) Viewed 3152 times

StarTools_2841.jpg (95.64 KiB) Viewed 3152 times

Also there’s no parameter to touch up your Apod Mask if Stars are missed or half covered ( this feature is only in a standard Mask where you can use Grow or Shrink etc…. )

Hmmm... Did you perhaps miss the Mask button?

As for 1.7 synthetic decon being better than 1.8, you will (hopefully) find that you are mistaken, particularly when employing it for planetary or solar imaging (where star samples are unavailable). You should find 1.8 yields superior detail and can be pushed more.

Hope this helps,

[Startrek]

Hi Ivo,
Thanks for your reply and explaining the relevant points
I had no idea that you could use the Stardard Mask button and it’s editing controls ( grow / shrink etc… ) to touch up an Apod Mask

Well leave comparisons alone for the time being but I have always said the 1.8 SV Decon Sampled provides superior Deconvolution to that of 1.7 Decon , it was just my Stars that suffered due to Binning

I just completed a reprocess of M8 Lagoon but this time without Binning and SV Decon Sampled method
I used the standard Mask editing controls to touch up the Apod Mask ( clicked on some stars and used Grow )

The result was excellent, with stars looking exactly how I want them to look and Deconvolution really making the image of a higher standard with fine detail

How I manage Binning in the future I’ll just have to think about it on case by case scenario
Without Binning processing does take considerably longer but I’m happy with these results

Thanks again

Clear Skies
Martin

[admin]

That's fantastic to hear Martin
(and a bit of relief )

[Mike in Rancho]

Thanks Ivo and Martin.

Well, my stars are still coming out of SVD often mangled, so after reading through this thread again I went to do some experimenting tonight --

I think I've just been binning too much for my scale. With APS-C, so over 6000x4000, and a pixel scale somewhere around 1.3", I've always figured that binning to 50% or 35%, putting me in the 2200 pixel wide range, would be enough resolution for deconvolution.

I brought in some mono Ha data on the Wizard and did just that, as usual, but no matter what I tried in many cases I just couldn't shake the artifacts, wild pixels, and oddly blocky star cores. So I got to ignoring them or fixing them at the end (often only really seen at 200 or 400% scale anyway).

Now I know that I do saturate a few star cores, but really we almost all do that and I figure I'm fairly conservative that way. Out of 26MP or whatever, I set my subs so that NINA says maybe 100 to 300 (depending) are maxed out. Trying not to blow out any pixels would be almost futile. 6" isn't that much aperture, but it is f/3.8 with the CC.

So I re-did it but no binning, kind of what I think Martin was trying out above. That seemed to work significantly better, although the SNR was poor and the computer time was longer. All I had done was my crop, AutoDev, quick Contrast, and then to SVD. After that, I binned down to my more usual 2200 wide to get the SNR, and the stars looked quite good.

So, at least with my funky data, it seems that I almost have to go back to a legacy ST workflow of SVD near the beginning.

I know that may harm some of SVD's time traveling features of taking into account what happens in the other modules, but...I don't think I have a choice.

I will have to experiment or think further on which, if any, of the other modules I might want to have applied at either the pre or post bin stage, in doing it this way, and if pre bin whether to do pre or post SVD.

I also still need to figure out just what AutoDev (or other modules) help to get me the nice stars with fat green cores, rather than the big red blobs with a tiny green or yellow dot. Or green but with red or yellow in the center. So many stilly stars I seem to have.

[Startrek]

Mike,
Yes I think my testing has confirmed that Binned data just doesn’t allow SV Decon to provide a satisfactory result with your Stars
It’s really a catch 22 , your data is a bit noisy so you Bin it to achieve better SNR ( which works well ) at the expense of some reduction in resolution. Then you want to use Deconvolution to improve the fine detail but your stars end up suffering
SV Decon does work extremely well but only when your dataset allows it to.
So in my case as long as my data is good ( ie minimal noise ) I can process Un binned and SV Decon Sampled method will work and work well.
The problem is a lot of my captures are in Bortle 8 with a OSC and noise is always an issue
Maybe I’ll have to reluctantly make the jump to Mono sooner than later as without a doubt this is the best way to image under any sky condition in both Broadband or Narrowband

In regard to Star samples I to have trouble finding good samples with green cores , mine are mostly faint green or yellow ( most of the stars with large green cores are located in nebulosity) I’m lucky to find day 8 good samples in the last few data sets

Anyway it’s been certainly a learning curve
Ivo has been so helpful and also everyone else who has posted on this topic

Clear Skies
Martin

[hixx]

Hi Martin, Mike,
to create a good PSF model it's most important to cover all regions in the image rather than choosing green stars only. If You see few greens only, have You tried
- different exposure
- excluding focus problems
- improve polar aligmment
- use guiding
- excluding stacking issues

Having said this, I am* using an 4.2" f/5.1 APO with just 550mmm focal length and a full-frame 24 Mpx Sony A7-III, so there will be tons of stars in that field.... I use to end up with loads of green stars, especially in busy star fields.
When running out of greens,You might want to pick a yellow or two sitting in that area. You may chose 2 or 3 yellows, to keep them from introducing the same error into the PSF model. I use to cover the edges, the borders and the object area using 3-4 sample stars each. I think You could get away with less stars, but, that's how it works well for me.

Regarding the BIN vs noise problem:
I also need to shoot in a suburb area, and I went to throw more integration time at the image and optimized shooting:
First I make sure, the single subframe is shot noise limited, meaning I'm shooting @ optimal ISO/ gain, because shot noise can be mitigated by integration time. This area different for each cam. the shot noise limited area can be identified as the lowest ISO/ gain which is ISO invariant.
I.e.: Looking at the ISO vs read noise curve, there is a mostly linear region in the where doubling ISO means doubling read noise (decreasing 3 dB SNR). In this area it is irrelevant whether I crank up the ISO while capturing or let ST do the job in stretching - the noise will only be limited by shot noise in this scenario.I use to pick the lowest or second lowest ISO of that. For DSLRs You may review those curves on dxomark.com, For astro cams, you'd probably want to refer to the vendor.

From there I work ETTR (exposure to the right (of the histogram). I Make sure, the brightest stars are around 80- 90% so they don't overexpose. To further reduce Shot noise I shoot a Luminance stack using an IDAS NB-1 duoband filter, and an RRGB stack using a clear or UV/IR filter

Note:
I am using a tiny ioptron CEM25 mount, which carries my APO if there isn't too strong wind. It's guided by an MGEN-3 guider which is good for 60-120s subs @ ISO800 or 1600. Apart from satellite & plane shots, i am able to use around 90% of the subs.
So I am able to stack about 4 hours integration time (120-250 frames)in APP. This delivers great star renderings
for me with tons of greens in SVD.
This really helps to get rid of the ugly noise, so I don't need binning or get away with 71%. I rather Crop the widefield strongly
to reduce rendering time but retain resolution for SVD.

Clear Skies
Jochen

[Startrek]

Hi Jochen,
Thanks for your reply
Most helpful
I image at 900mm to 1250mm with 6” , 8” and 10” Newtonian reflectors
Mounts are EQ6-R pro and EQ8-R pro
My polar alignment is usually around 1 to 3 arc minutes
My PHD2 guiding on nights of good seeing is around 0.50 to 0.60 arc sec error total and average to poor seeing is 0.80
I usually have minimal image shift across 4 to 6 hours of imaging across the night sky even with dithering
So my DSS stacking usually works out with not minimal stacking artefacts
My camera is the ZWO 2600MC cooled OSC and is usually set to Gain 100 lowest read noise. Read noise at Gain 100 is 1.5e
The camera has a UV/ IR cut internal protective window
Camera has no amp glow
Dark current at -10C is just about nil
Dynamic range at Gain 100 is around 13.5 stops and well depth of 20,000
My image scale ranges from 0.62 to 0.86 arc sec per pixel depending on what scope I use so I’m well oversampled which is ideal for Binning to improve SNR at the expense of some loss in resolution, but at the other end not so good for SV Decon with respect to the look of my Stars.
So as you can see my set ups are geared up well for long exposure AP
I do have green core Star samples but they are usually found on a few larger stars and in amongst nebulosity
It’s only early days with SV Decon so I have a lot of imaging to do before I can really assess it’s performance but so far Binning my image has a negative outcome in relation to Stars only ( the atmospheric blur reduction works fine , no issue ) it’s just the Stars.

I’m sure Mike may have some comments to make as well in relation to Binning, Star Samples and SV Decon in general

Thanks again
Martin

[Mike in Rancho]

Hi Jochen and Martin,

Me? Have comments? Yes I might do an offshoot thread as my thinking is now becoming more general about how to fit in an unbinned SVD in order to "save" my star shapes. It is related to and prompted by the discussion here, but perhaps veers away from the exact topic.

More on point to the green cores, and perhaps Ivo's help might be needed here, but, what exactly do colors we are seeing in the sample mask mean?

The nice fat green center, surrounded by a ring of yellow, then red, and the white outline, are easy to pick out but sometimes few and far between. I know from the guidance that it is more important to pick a yellow core sample in order to get coverage in various spatial regions.

But what makes good/bad samples beyond the easy picks like those? And what about the white line? Is SVD actually using the apod white line as the profile, or is it digging deeper into the actual color variations/profile within the while circle? Well, one hopes it is pretty circular, anyway.

Going further, what about stars whose center isn't so fat but rather just a red star, circled in white, but a small yellow or green spot in the center? No bueno? Or okay?

Likewise, we are told not to pick a green-cored star with red in the center - I presume these might be blown out pixels? Unsure, because we also have red in the outer stellar profile and those pixels are clearly not oversaturated. Also, sometimes an otherwise seemingly good green cored star might not have a spot of red in the center of the green area (don't use), but yellow. Does the same concept apply to those, or can they be used because no center pixels have gone fully to red? i.e, is it still ripe, or rotten?

I too, at least with the 2600, only use the two main "good" gains of 0 and 100, actually typically 0, just to have the bigger well, even in narrowband. I use NINA's statistics screen to make sure I have limited pixels at max saturation. However, it seems to be also that the level of binning perhaps, and final AutoDev, have a great deal to do with how the colors in the samples turn out. But I haven't really experimented to figure out a pattern yet. We also do have a linearity control in SVD, but likewise I am unsure if that is something that should be fiddled with in the normal course of things.

[admin]

SV Decon should definitely still be able to yield good results and meaningful improvements on binned data. Even for data that is undersampled. The only issue you may run into, is that you may require quite a few more samples (if using sampling), as reconstructing PSFs from undersampled data is a lot harder (providing more samples essentially has the same effect of "drizzling").

hcrop_notsvd.png (208.36 KiB) Viewed 3061 times

hcrop_svd.png (249.17 KiB) Viewed 3061 times

(processed at actual 763x551 resolution)

If you have any datasets that are troublesome, please feel free to share them with me.

More on point to the green cores, and perhaps Ivo's help might be needed here, but, what exactly do colors we are seeing in the sample mask mean?

The docs say that "if you cannot find such stars and you need samples in a specific area you may choose samples that have a yellow core instead. As a rule of thumb, providing samples in all areas of the image takes precedence over the quality of the samples."

The colors give an indication of the quality of the signal (linearity, SNR and gradient slope). The better the signal, the less noise the sample will have and the better (more cleanly) it will describe the PSF.

Hope that helps!

[Mike in Rancho]

The docs say that "if you cannot find such stars and you need samples in a specific area you may choose samples that have a yellow core instead. As a rule of thumb, providing samples in all areas of the image takes precedence over the quality of the samples."

The colors give an indication of the quality of the signal (linearity, SNR and gradient slope). The better the signal, the less noise the sample will have and the better (more cleanly) it will describe the PSF.

Hope that helps!

Thanks Ivo. Yes I've read that in the docs and even quoted it (loosely) in my post. I'm just trying to dig deeper into the apod mask colors - and perhaps am looking for meaning that isn't there. Wouldn't be the first time.

I may have to try your suggestion above that, with undersampled or perhaps more binned-down sets, to use many samples to make up for that. I'll start testing that out.

Possibly partially related to the other thread's topic that I need to work on also, sometimes it seems I see different quality coloring in the apod mask even with the same underlying dataset, when reprocessing from scratch. That makes me wonder if perhaps binning, AutoDev, and/or module work before SVD is incorporated into that quality evaluation, and perhaps I should be more mindful of what I am doing in order to "help" the eventual use of SVD. As always, I must do more testing.

These clips aren't from the greatest data, and also from my Newt which seems to profile the stars a bit differently than when I load up refractor data --

star sample strip.jpg (90.3 KiB) Viewed 3024 times

Panel 1 is your basic red ball. I assume these are not to be used for SVD (I've never tried, actually). Problem is, sometimes my entire apod mask is pretty much only stars just like this.

Panel 2 has some yellow in the core, no green. Ostensibly then okay for a secondary choice if I have nothing else in the area?

Panel 3 has some green. Not a ton, but pretty close to the best that I typically see. Assume this should be selected.

Panel 4, (if one ignores the adjoining star), is green but red pixeled in the center. Not to be used even if it was by itself.

Panel 5, (again ignore the double star for now), is starting to show a green center but there's a yellow pixel (though not gone red) in the very center. Still okay? I think the Features and Docs examples of green stars actually may have a central yellow dot themselves, if I am peeking at it correctly?

So, other than maybe binning or stretching differently (I can always redo AutoDev I suppose) before SVD, can anything else be done to help out the population of star samples?

The only setting that seems to affect the apod quality colors is Linearity. Now, it says this is to be used in case there's something funky with your sensor, and I'm pretty sure my sensors are normal. That said, moving the slider to the right seems to tamp down or even eliminate red pixeled cores, while moving the slider to the left can help "find" yellow or green cores in otherwise red ball stars (whilst pushing other stars into red pixeled cores, if any are prone to that). Can that slider be used to find a better overall sample quality balance to the apod mask, or would that be a false usage and it really is only for unique sensor response?