Jump to content

  •  

CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.

Photo

Drizzle benefits question

  • Please log in to reply
53 replies to this topic

#1 Gschnettler

Gschnettler

    Apollo

  • *****
  • topic starter
  • Posts: 1,045
  • Joined: 14 Jul 2020
  • Loc: Cincinnati Ohio

Posted 29 May 2024 - 10:47 AM

I am wondering if there is a way to figure out how much benefit drizzling provides. For example, let’s say that someone has a Tak FSQ-106EDX telescope and a ZWO ASI 2400mc camera and their image scale is 2.31”. Let’s say they live somewhere with reasonably good seeing and they are undersampled. So they try 2x drizzle and they see an improvement in resolution.

Now let’s say that they have a friend who lives nearby who has a Tak FSQ106EDX and a ZWO ASI 6200mc pro and their image scale is 1.46”. They are not undersampled and thus they decide not to drizzle.

Now these two imagers get together and do a test by imaging the same target on the same night for the same amount of time keeping everything consistent between their two set ups as much as possible so the only difference is with the cameras and the dithering.

They both hand their data off to a friend who is good at processing and he applies the same workflow and techniques to both data sets, the only difference being 2X drizzle on the 2400mc pro data.

Can we anticipate how the 2 resulting images would compare? Would they basically be the same or would we expect one to be better than the other in terms of signal, noise, resolution, etc?

#2 grazer

grazer

    Lift Off

  • -----
  • Posts: 3
  • Joined: 04 Jan 2019

Posted 29 May 2024 - 11:07 AM

Hi,

 

coincidentally, I saw this video recently.  https://www.youtube....h?v=IlQb_23Kls4 The chap followed it up with two more, I think you MIGHT like them.



#3 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 34,863
  • Joined: 27 Oct 2014

Posted 29 May 2024 - 11:19 AM

They'd be pretty similar.  Drizzling is a tradeoff, trading snr for (maybe) resolution. But image scale also affects snr.

 

Many things factor into this.  Seeing is a big one.  Looking at the situation (overly) simplistically, 2.31 drizzle is sort kinda like 1.16.  So, compared to 1.46, there is POTENTIALLY a bit more resolution, and a bit worse snr.

 

But the reality of the situation is that other factors will intrude.  To include everything, you MUST actually do the experiment.

 

There's a reason why that youtube says "we discover that theory often doesn't hold up to practical tests".  Which is exactly what I'd expect.

 

This is a real problem on CN.  Often, someone will state something as a certainty, based on a theoretical analysis, when in reality the results of that analysis are totally driven by what factors they include and what they don't, and therefore pretty meaningless. 

 

Real tests are MUCH superior.

 

But even they are not definitive.  Note that here, the results of the test will be affected by (among other things) what the seeing is on that particular night.  On another night they might change.

 

Complicated business, this.  The choice between these two things will be different for different imagers.

 

I'd expect the results would be similar.  Personally I'd use the smaller pixels of the 6200 and skip the drizzle.  I prefer setups with better snr.  My total imaging time could be a bit less that way.

 

YMMV.  <smile>


Edited by bobzeq25, 29 May 2024 - 11:29 AM.

  • RedLionNJ, limeyx, Skysmacker and 2 others like this

#4 Gschnettler

Gschnettler

    Apollo

  • *****
  • topic starter
  • Posts: 1,045
  • Joined: 14 Jul 2020
  • Loc: Cincinnati Ohio

Posted 29 May 2024 - 11:38 AM

Thanks.

How do you get from 2.31” image scale to 1.16 with 2X drizzle? Is there a formula for that?

I agree that real world tests are the best.

I’ll watch the video when I get some time after work.

#5 Oort Cloud

Oort Cloud

    Fly Me to the Moon

  • -----
  • Posts: 5,581
  • Joined: 19 Nov 2020
  • Loc: New Jersey, USA

Posted 29 May 2024 - 11:52 AM

Thanks.

How do you get from 2.31” image scale to 1.16 with 2X drizzle? Is there a formula for that?

I agree that real world tests are the best.

I’ll watch the video when I get some time after work.

He's assuming a 2x drizzle, so ½ the scale. Some software will also let you do a 3x drizzle, in which case it would yield 0.77333" per pixel.

3x is almost always overkill, resulting in enormous files with little to no advantage over 2x (extra noise with no extra detail, so actually a detriment).


Edited by Oort Cloud, 29 May 2024 - 11:55 AM.


#6 auroraTDunn

auroraTDunn

    Viking 1

  • -----
  • Posts: 715
  • Joined: 01 Sep 2017
  • Loc: North of Boulder, Co.

Posted 29 May 2024 - 12:04 PM

As noted above Sky Story, on YT, has a few brand new videos discussing this. Lukomatico has a couple good ones too.

 

In the end I always do a 1x drizzle and then check the drizzled vs non-drizzled out closely in PI before deciding which to use. Often it does not matter unless I'm intentionally setting my subs times such that my end result sampling suggests drizzling will help. The other reason why, I personally, use 2x drizzle is if I really want a small tight crop of something because Im using, most of the time now, a RC71 (widefield) + a ASI6200MM which has a large sensor so can handle a tight 2x drizzle (for instance, galaxies and globs).

Make sure you have at least 15 frames, per filter if filtering but the more the better as drizzling injects noise. Since I usually do deep/long imaging with 300sec subs (the vast majority of time) I don't worry about the increased noise as NXT deals with it very well!

 

In the end I would suggest experimenting with your images with no, 1x and 2x drizzling and find what works best for your desires.

 

 

How do you get from 2.31” image scale to 1.16 with 2X drizzle? Is there a formula for that?

For this i always use Topaz Gigapixel or Topaz PhotoAI. You could always just do a resample in PI or app of use or use something like PS, LR or whatever your using but I rarely us them as they, currently, do a **** poor job, relatively speaking, to Topaz. There are free AI resizers out there which can be found but their models include no astroimages so user beware.



#7 Oort Cloud

Oort Cloud

    Fly Me to the Moon

  • -----
  • Posts: 5,581
  • Joined: 19 Nov 2020
  • Loc: New Jersey, USA

Posted 29 May 2024 - 12:37 PM

...

Often it does not matter unless I'm intentionally setting my subs times such that my end result sampling suggests drizzling will help.

...


Just FYI, drizzling and sampling has zero to do with exposure length.

If you were undersampled for the seeing, and you dithered your lights, then drizzling can add resolution at the cost of also adding noise. It will always add noise, whether it adds detail will depend on the two conditions above being met.

That's really all there is to it. The only way sub-length factors in is if you used subs that were long enough that you didn't collect a lot of them. Then drizzling would be inadvisable as it requires large datasets to truly be effective.

#8 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 34,863
  • Joined: 27 Oct 2014

Posted 29 May 2024 - 02:03 PM

Thanks.

How do you get from 2.31” image scale to 1.16 with 2X drizzle? Is there a formula for that?
 

Rough idea.  2X drizzle is sorta kinda like dividing your image scale by 2.

 

Note that I said (twice) sorta kinda.  <smile>  That "other factors count" thing.
 


Edited by bobzeq25, 29 May 2024 - 02:04 PM.


#9 auroraTDunn

auroraTDunn

    Viking 1

  • -----
  • Posts: 715
  • Joined: 01 Sep 2017
  • Loc: North of Boulder, Co.

Posted 29 May 2024 - 02:44 PM

Just FYI, drizzling and sampling has zero to do with exposure length.

If you were undersampled for the seeing, and you dithered your lights, then drizzling can add resolution at the cost of also adding noise. It will always add noise, whether it adds detail will depend on the two conditions above being met.

That's really all there is to it. The only way sub-length factors in is if you used subs that were long enough that you didn't collect a lot of them. Then drizzling would be inadvisable as it requires large datasets to truly be effective.

That's complete true. The issue I have, again with my personal workflows, is if I'm well to oversampled (most of the time this is the case) and I'm taking long exposures for faint aspects but have dense and/or other bright areas life becomes VERY hard time trying to get a good nonlinear stretch which is made worse(OK harder not worse) with 2x drizzling. But I'm spending more time playing with HDRMT, thanks to Adams videos and this really helps in these cases. Granted there should be no difference whether I drizzle or not when at that stage, yet I, again personally. always seem too.



#10 Marcelofig

Marcelofig

    Apollo

  • -----
  • Posts: 1,125
  • Joined: 21 Jan 2015

Posted 29 May 2024 - 04:17 PM

Recently this long and very technical thread about the use of drizzling was published in the PI forum. In case anyone here is interested.

 

https://pixinsight.c...eep-dive.23380/



#11 Spaceman 56

Spaceman 56

    Aurora

  • *****
  • Posts: 4,987
  • Joined: 02 Jan 2022
  • Loc: New Zealand

Posted 31 May 2024 - 05:38 PM

I read that Drizzling is completely pointless if you forgot to Dither.  

 

experts please advise if this is correct or not.  smile.gif

 

Spaceman



#12 idclimber

idclimber

    Fly Me to the Moon

  • *****
  • Posts: 6,698
  • Joined: 08 Apr 2016
  • Loc: McCall Idaho

Posted 31 May 2024 - 05:45 PM

I read that Drizzling is completely pointless if you forgot to Dither.  

 

experts please advise if this is correct or not.  smile.gif

 

Spaceman

Yep, you need sufficiently dithered data for it to work. 

 

Earlier this spring I did some testing on some old data. I drizzle integrated a bunch of lum data that I had both with 1x and 2x. The 2x data was subsequently downsampled by a factor of 2 to return to the native image scale. This gave me superior detail to the non drizzled data and the 1x. It also had the same noise.


  • limeyx, Spaceman 56 and soojooko like this

#13 danny1976

danny1976

    Surveyor 1

  • -----
  • Posts: 1,849
  • Joined: 02 Nov 2021
  • Loc: Belgium

Posted 31 May 2024 - 06:23 PM

Correct me if I’m wrong, but dithering+drizzling reminds me of pixel shift on some mirrorless cameras like the Sony A7RV. Here the camera sensor is shifted a few pixels by the stabilization instead. 



#14 Gschnettler

Gschnettler

    Apollo

  • *****
  • topic starter
  • Posts: 1,045
  • Joined: 14 Jul 2020
  • Loc: Cincinnati Ohio

Posted 01 June 2024 - 06:20 AM

This is all good information.

How much dithering do you have to do in order for it to be sufficiently dithered?

#15 Spaceman 56

Spaceman 56

    Aurora

  • *****
  • Posts: 4,987
  • Joined: 02 Jan 2022
  • Loc: New Zealand

Posted 02 June 2024 - 04:06 AM

This is all good information.

How much dithering do you have to do in order for it to be sufficiently dithered?

the idea is to move the pixel position between sub exposures.

 

some people dither every frame, but many will shoot 5 or 6 frames before dithering. 

 

sometimes I dither every 10 frames, as settling time can eat into imaging times, but I think every 5 or 6 frames is better.

 

Spaceman



#16 GiffS

GiffS

    Viking 1

  • -----
  • Posts: 518
  • Joined: 20 Nov 2021
  • Loc: SW Virginia

Posted 02 June 2024 - 09:24 AM

This is all good information.

How much dithering do you have to do in order for it to be sufficiently dithered?

I have been using a frequency of every 2 frames and a 2 pixel move recently. Like Spaceman I used to use every 5 or 6 subs with the AVX mount due to settling times but with the AM5 it is no longer an issue. I am just experimenting with drizzling in PI and I have much to learn.

 

For example it seems to me that the drizzled output at 2X drizzle requires extreme care in post processing with respect to noise. It seems to be very easy to create a very harsh noise in the background. View Into Space did a bit on one of his YT videos where he created a Range Mask that protects the background while you stretch the stars and galaxy elements and that has helped. I have also dropped to 1X unless I expect to crop aggressively on a small portion of the image and on the same data that seems to reduce the noise. Does that make sense?

 

One thing that is probably dead simple, but I am over thinking it, has to do with down sampling a drizzled image. I've seen it mentioned that the real magic happens when you downsize or down sample the drizzled image. Is that just a matter of resizing the image in a normal graphics application or is there some other process in use? Sorry if that's a stupid question.



#17 imtl

imtl

    Cosmos

  • *****
  • Posts: 8,410
  • Joined: 07 Jun 2016
  • Loc: Down in a hole

Posted 02 June 2024 - 09:50 AM

That's complete true. The issue I have, again with my personal workflows, is if I'm well to oversampled (most of the time this is the case) and I'm taking long exposures for faint aspects but have dense and/or other bright areas life becomes VERY hard time trying to get a good nonlinear stretch which is made worse(OK harder not worse) with 2x drizzling. But I'm spending more time playing with HDRMT, thanks to Adams videos and this really helps in these cases. Granted there should be no difference whether I drizzle or not when at that stage, yet I, again personally. always seem too.


You should not drizzle when you are oversampled. You're just reducing your SNR even more and gaining absolutely no new information.
  • RedLionNJ and Oort Cloud like this

#18 imtl

imtl

    Cosmos

  • *****
  • Posts: 8,410
  • Joined: 07 Jun 2016
  • Loc: Down in a hole

Posted 02 June 2024 - 09:56 AM

the idea is to move the pixel position between sub exposures.

some people dither every frame, but many will shoot 5 or 6 frames before dithering.

sometimes I dither every 10 frames, as settling time can eat into imaging times, but I think every 5 or 6 frames is better.

Spaceman

The idea is to RANDOMLY shift the frames around.

Dithering every 10 frames is quite useless. Especially with your OSC since you probably bayer-drizzle your lights. You should drizzle a lot more often.

People tend to focus only on ''getting most frames'' in a given time. The goal in acqusition is to get to gain the most useful frames to eventually get the best master light to process. That includes dithering, well matched calibration and near perfect focusing.

Avoiding dithering a lot because ''settling time is wasting acquisition time'' is the wrong way of doing things. Subs don't matter. Master light does.

Edited by imtl, 02 June 2024 - 09:56 AM.

  • pedxing, sn2006gy, Phishin_phool and 2 others like this

#19 Der_Pit

Der_Pit

    Mercury-Atlas

  • *****
  • Posts: 2,893
  • Joined: 07 Jul 2018
  • Loc: La Palma

Posted 02 June 2024 - 12:34 PM

Dithering is to apply some statistical smoothing of (especially) fixed pattern noise.  For that it is crucial to have a good sample.  So the really important question is how many dither positions you will have in the end.  I always try to have at least 30-40 dither positions (per filter, as I do RGB).  I.e., the longer your subs are, the more often you should dither.

That way, I usually end up having some 10-15% time 'overhead', i.e., some 6-10 minutes per hour (in my case also includes the autofocus on filter change and/or temperature drift)


  • imtl likes this

#20 Spaceman 56

Spaceman 56

    Aurora

  • *****
  • Posts: 4,987
  • Joined: 02 Jan 2022
  • Loc: New Zealand

Posted 02 June 2024 - 07:05 PM

Dithering every 10 frames is quite useless. 

Why ?

 

and how often is required ?



#21 acrh2

acrh2

    Mercury-Atlas

  • *****
  • Posts: 2,817
  • Joined: 16 Mar 2021

Posted 02 June 2024 - 07:11 PM

Dithering every 10 frames is quite useless. Especially with your OSC since you probably bayer-drizzle your lights. You should drizzle a lot more often.
 

 

What happens if you have 1000 frames in a stack? Then dithering every 10 frames seems to be quite sufficient. 


  • Der_Pit and limeyx like this

#22 DanMiller

DanMiller

    Soyuz

  • -----
  • Posts: 3,966
  • Joined: 20 Jun 2022
  • Loc: Pennsylvania, USA

Posted 02 June 2024 - 07:23 PM

Every 5 miuntes, give or take dependeing on exposures. If it is 4 minute exposures, dither every frame. If it is 120s exposures, dither every 3 frames.  Why choose frames, choose a time period.  If youy have 8 hours of imaging at 120s, you have 240 frames.  Dither every 3 frames, which is every 6 minutes of imaging gives you 40 frames of dithering.    16 percent of your frames are dithered.  

 

Actually, lets ask the question.  What percentage of your time should be dithered?  Every frame is 100 percent.  Who here actually dithers every frame, or at 100 percent unless they are doing 5 minute exposures.

 

DAn


  • limeyx likes this

#23 imtl

imtl

    Cosmos

  • *****
  • Posts: 8,410
  • Joined: 07 Jun 2016
  • Loc: Down in a hole

Posted 02 June 2024 - 11:51 PM

What happens if you have 1000 frames in a stack? Then dithering every 10 frames seems to be quite sufficient.


In practice? Perhaps. But Spaceman is not acquiring 1000 frames in a stack. Also, what statistics did you use to decide that every 10 is enough dithered frames in a stack of a 1000? Keep in mind I was referring to Bayer drizzle.

#24 imtl

imtl

    Cosmos

  • *****
  • Posts: 8,410
  • Joined: 07 Jun 2016
  • Loc: Down in a hole

Posted 02 June 2024 - 11:53 PM

Every 5 miuntes, give or take dependeing on exposures. If it is 4 minute exposures, dither every frame. If it is 120s exposures, dither every 3 frames. Why choose frames, choose a time period. If youy have 8 hours of imaging at 120s, you have 240 frames. Dither every 3 frames, which is every 6 minutes of imaging gives you 40 frames of dithering. 16 percent of your frames are dithered.

Actually, lets ask the question. What percentage of your time should be dithered? Every frame is 100 percent. Who here actually dithers every frame, or at 100 percent unless they are doing 5 minute exposures.

DAn


Because the whole point of dithering is to help randomize FPN in a registered stack. Frames are what's important. Not time. The only reason people don't dither every frames is because of wanting to maximize acquisition time. Which is a fair goal of course.

#25 acrh2

acrh2

    Mercury-Atlas

  • *****
  • Posts: 2,817
  • Joined: 16 Mar 2021

Posted 03 June 2024 - 12:09 AM

In practice? Perhaps. But Spaceman is not acquiring 1000 frames in a stack. Also, what statistics did you use to decide that every 10 is enough dithered frames in a stack of a 1000? Keep in mind I was referring to Bayer drizzle.

 

You need ~25 dithers per stack to fully sample the Bayer matrix. I am going to assume the same for sampling x2 drizzle pixels (same 2x2 layout.) Averaging 10 undithered frames is exactly like having a single frame with sqrt(10) less noise. So a 1000 frame stack with dithers for every10th frame is equivalent to having a stack of 100 frames with dithers for every frame. Does that sound right?




CNers have asked about a donation box for Cloudy Nights over the years, so here you go. Donation is not required by any means, so please enjoy your stay.


Recent Topics






Cloudy Nights LLC
Cloudy Nights Sponsor: Astronomics