Lucky imaging - stacking faint stars

[Ags]

I was shooting M13 and M57 last night and I got promising but frustrating results. I was shooting with the following parameters:

Scope: C6 @ F10 on AZ-GTi
Camera: ZWO ASI 178 MM (non-cooled)
Sequence: 500 x 333 ms subs
Gain: 310
Binning: 2x2
ROI: full chip

I processed in AS3! using 1.5 drizzle with Analyze set to focus on very small / high SNR. Why bin 2x2 then drizzle? Binning is to try help AS3 as with short DSO subs it can struggle to align the frames with all the noise. Drizzle then tries to get the detail back and unsquare the stars. Then I did a quick stretch in Gimp:

I know a three minute sequence is not nearly enough but I have shot several - so the above is really just one "sub" with about 100 seconds of exposure (after 40% of the frames were discarded). Overall I am happy with the above, with the glaring exception of faint stars. The AS3! stacking seems to fail to align the faint stars resulting in any star below a threshold showing as a blob. My question is: how can I persuade AS3! to clean up the faint stars?

 

 

 

[Ags]

Anyone?

[andrew s]

Not an expert or ever mildly qualified in this area but how can you align on some stars but not others? Is it because there there are distortions between frames?  In which case you need to reject the poor ones.

Regards Andrew 

[Ags]

I am rejecting frames - in fact I am throwing away nearly half of them. Each frame is covered in a web of independent alignment points - and it seems if there is not enough signal from a star its local alignment point doesn't work.

Here is my attempt at M57 from the same session. Focus was not the best on this one!

[andrew s]

Sorry can't help out of my league! Regards Andrew 

[Ags]

Me too!

[CraigT82]

Can you do a slight pre-stretch on each sub before stacking? Batch process somehow?

[Ags]

Guess I could - but I am very close to the noise floor - so I may end up with the same problem just with bigger numbers. I do kind of do that as my camera is 14 bit but the data is so epicly dim I process it as 10 bit data, so I am stretching by a factor of 16. I am hoping an incoming 6.3 reducer will help but I wouldn't be using it with M57 anyway as it is tiny and bright.

Edited May 16, 2020 by Ags

[happy-kat]

Is this experience across all targets or just targets similar to the two here?

[Ags]

I see it with all DSO targets. I suppose I could try longer subs.

[happy-kat]

could look at a stacked comparison, like use DSS or Siril.

500 x 333 ms subs = is that not just effectively just under 3 minutes of exposure total, that is not long and you got an image with clarity to see what the target was

Edited May 16, 2020 by happy-kat
not enough 0

[Ags]

I figured it out. There is a setting to apply alignment globally and not align every alignment point individually. 

I never understood why you would want to set Global, this is why - if you align at frame level then dim stuff stacks reliably. Use Local for the Moon, where there is always plently of local data.

[happy-kat]

I still think it's quite incredible to get the image you had for less than 3 minutes of total exposure.

[Ags]

Thanks @happy-kat. I am pleased too with the result, particularly at getting this definition at 1500 mm focal length on an AZ-GTi.

Here is the results of using the global setting. Both are just quick stretches in Gimp.

Before (using local alignment)

After (Using Global alignment):

1. 

 

Definitely more definition in the faint stars around the cluster. 

Edited May 17, 2020 by Ags

[happy-kat]

Plus you are using the mount in az mode I think.

Yes I see more definition to and that's on a phone that has display algorithms that eat stars.

[Ags]

Yes AZ mode - that is what limits my sequences to a couple of minutes. With a wedge I could be shooting 10 minute subs. Actually I could shoot subs of any length I liked - 1 hour would not be an issue, except for the fact I have to watch the screen and keep the target on the crosshairs! Maybe tracking will be better in EQ mode though...? My main issue with tracking is the ALT slips a gear tooth every minute - I guess in EQ mode my buggy ALT gear wouldn't be spinning so no problem.

Edited May 17, 2020 by Ags

[happy-kat]

Use the head of your az3 as the wedge.

If guiding then both axis are still used but in eq tracking only the one axis is used I gather.

[Ags]

I was just regretting selling that AZ3...!

[vlaiv]

On 15/05/2020 at 20:28, Ags said:

I processed in AS3! using 1.5 drizzle with Analyze set to focus on very small / high SNR. Why bin 2x2 then drizzle? Binning is to try help AS3 as with short DSO subs it can struggle to align the frames with all the noise. Drizzle then tries to get the detail back and unsquare the stars. Then I did a quick stretch in Gimp

That won't work.

12 hours ago, Ags said:

I figured it out. There is a setting to apply alignment globally and not align every alignment point individually. 

You need to know what that option does - and in fact it is pretty clear in your example what the difference is.

It is related to quality estimator and stacking. Global will accept / reject whole frame based on quality threshold. Local will accept / reject local area around alignment point (and related square) - thus taking only a portion of sub for a stack - one that is sharp.

With global you will stack full frames and with local you will stack pieces of each sub that have good enough quality.

Global:

- uniform SNR improvement

- all stars will be equally tight (both those that have alignment points and those that don't)

Local:

- varying SNR improvement unless you take very large number of subs and things average out (there is on average same number of good patches all over the image)

- you need to cover whole image with alignment points for this to work properly and there needs to be a "feature" around each alignment point that is used for aligning. Empty space creates a point - this is best done on Lunar surface and such

- very different tightness of stars - where you have alignment point, you'll have tight stars and good SNR - where you don't have one - stars will be smeared.

I would say that best approach for Lucky imaging with small aperture is to do global rejection of subs and then take model that only includes translation and rotation for aligning subs.

AS!3 uses local geometry adjustment and I believe that this can be problem in your case. This sort of aligning and stacking deforms each sub to correct for motion of atmosphere. Such motion happens on very short scales - 10ms and less.

You are not imaging at such short scales -  and that would be next to impossible with 6" of aperture. You might be even using shorter exposure than it is really needed for what you are doing. There is very small difference between 1/3s, 1/2s and 1s exposures for what you are trying to do.

Atmospheric movement and distortion happens at scales of 5-6ms. 1/3s exposure will have about 60 of such intervals - things are bound to average out (1/2 will have 100 and 1s will have 200 - not much of a difference - each one will average out). For this reason you can't try to do what planetary imaging does - use even such deformed frames by correcting them back.

Only thing that you can do - is select subs that were shot in periods of more stable seeing vs periods of poorer seeing. That and avoid any issues with mount smoothness and guiding.

One could in principle write algorithm to use local geometry correction in this case as well - using alignment points - and even alignment points for stars that have poor SNR in single sub - but it would take ages to stack such image. Algorithm would first stack every frame to get good enough SNR to detect each star in the image - and then try go guess where that star is in each individual sub to do local geometry correction.

 

[Ags]

Yes I figured out i need to use Global. I see Emil Kraaikamp uses frames of 1 second and still calls his results lucky imaging. I would go longer than 1/3 or 1/2 second, but the limiting factor is for me is tracking not seeing. The 6.3 reducer should let me add a few more milliseconds. 

Why won't my drizzle trick work?

[vlaiv]

26 minutes ago, Ags said:

Yes I figured out i need to use Global. I see Emil Kraaikamp uses frames of 1 second and still calls his results lucky imaging. I would go longer than 1/3 or 1/2 second, but the limiting factor is for me is tracking not seeing. The 6.3 reducer should let me add a few more milliseconds. 

Why won't my drizzle trick work?

Drizzle needs very specific set of conditions to provide benefit. To name a few:

- undersampling

- precise dither offset

In most cases these two are not met in amateur setups and drizzle does not produce wanted results - resolution improvement. Often, smooth stars are being reported as reason to use drizzle - but that can be accomplished by using a bit more sophisticated resampling algorithm - you don't need to drizzle. On the other hand, when drizzling - you are spreading around your samples and lowering SNR.

Although it is widely available and often used - I would say that drizzle is one of the least useful tools astro imagers have at their disposal and often misused.

[happy-kat]

Drizzle though it surely helps by minimising walking noise which is dispersed when drizzling.

EDIT ignore the above I'm thinking about dithering NOT drizzle

Edited May 17, 2020 by happy-kat

[vlaiv]

9 minutes ago, happy-kat said:

Drizzle though it surely helps by minimising walking noise which is dispersed when drizzling.

Maybe you have mixed dithering and drizzling?

Dithering will sort out walking noise (in conjunction with other algorithms like sigma clip and so on).

[happy-kat]

Gosh yes you are right the Ds swapped I was thinking dithering but read drizzle.

[Ags]

Here is my latest version of M57 - 17 2 minute sequences of 240 frames (frame length being 500 ms clearly). I kept 40% of the frames. I think (after a bit of sharpening to see what's there) I am starting to pick out some detail in the nebula. Still very noisy but I think I can add to this. Also need to work on not overexposing the stars - this happens when i linear stretch the data for DSS (otherwise the stars are too faint to register). I think I can improve here though. @vlaiv - no drizzle in this version.

Edited May 17, 2020 by Ags