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

Dithering - Drizzle - Resolution

  • Please log in to reply
16 replies to this topic

#1 Benschop

Benschop

    Viking 1

  • *****
  • topic starter
  • Posts: 508
  • Joined: 12 Jan 2015
  • Loc: Ontario

Posted 25 October 2021 - 10:05 PM

Cloudy / Rainy night question...

 

With my previous ED80 at 600mm focal length, resolution was 1.85"/pixel and I never had "blocky" stars.

With my new widefield scope (GT71), imaging camera resolution without flattener/reducer is 2.65"/pixel, with the Flat 6A iii flattener/reducer, 3.31"/pixel.

I will generally be using the Flattener for all widefield imaging.

 

I am familiar with the "standard" image sampling guidelines (average seeing)... and I realize that the images will be undersampled at the above resolution.

Once things clear I will test this combination and if results are ugly, I will then migrate to a camera with smaller pixels.

 

The question I have while waiting for the sky to clear...

How much can one push the guidelines if dithering and drizzle stacking?

If I am to dither my guiding, and drizzle my stacking, can tight round stars be achieved with a resolution of 3.31"/pixel?

 

 

 

 

 

 



#2 licho52

licho52

    Mariner 2

  • -----
  • Posts: 222
  • Joined: 15 Dec 2020

Posted 25 October 2021 - 10:44 PM

Yes.


  • Benschop likes this

#3 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 26,271
  • Joined: 27 Oct 2014

Posted 25 October 2021 - 11:09 PM

Couple of things.

 

Note that increasing resolution with drizzle is not free.  You pay in lower signal to noise resolution.  You can compensate with more total imaging time.

 

How ugly is this?  2.8.  Better version than the blurry required CN jpg is referenced below.  I think people worry too much about this.  For me the important thing is that the image looks good full size.  Zooming in is virtually guaranteed to make you less happy.  I like happy.  <smile>

 

https://www.astrobin.com/367734/C/

 

NGC6992 HaO(III)RGB V4.jpg


Edited by bobzeq25, 25 October 2021 - 11:11 PM.

  • telfish and Benschop like this

#4 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 24,997
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 25 October 2021 - 11:10 PM

Drizzling will definitely improve the profile of your stars, and they should all end up more round. How round will depend on how well you've managed tilt in your system, but generally the blockiness of coarse resolution stars will be mitigated or eliminated with drizzling. Elongated stars won't magically become round with drizzling, but the edges will become softer and smoother. Round still requires good tilt and optical aberration management. 

 

As for tight, small....drizzling can potentially improve real-world resolution, with enough good quality data, and a more significant drop shrink. The drop shrink affects how spread out the drizzled information is actually drizzled. A drop shrink of 1.0 (at least in PixInsight) basically means no improvement in real-world resolution, 0.5 would potentially mean a 50% reduction in the size of details. Whether you can actually realize this potential, however, depends on the data...if the source data is soft with bloated details, drizzling, even with a significant drop shrink like 0.5, will not actually change much, if anything. If you have very crisp, sharp, detailed source data, then drop shrink will usually be able to eek out greate rimprovements in resolution.

 

That said, its not a magic bullet, and it really depends a lot on the data and kind of details you are drizzling. In my experience, a drop shrink of 0.5 with very good data, produces a notable but not hugely significant improvement in the fineness of the details I was able to capture. I also find that drizzling, performing careful and moderate deconvolution on the full size drizzled image to tighten things up just a bit more, then downsampling back to the original native camera resolution, nets the best results. You are really able to see the improvements that drizzling and deconvolution can deliver when you do that.

 

JjVm5kd.gif


  • Dennis_Oz, rgsalinger, jacquesdeacon and 9 others like this

#5 Benschop

Benschop

    Viking 1

  • *****
  • topic starter
  • Posts: 508
  • Joined: 12 Jan 2015
  • Loc: Ontario

Posted 25 October 2021 - 11:23 PM

Bob - good point on the SNR - had not considered that. But at least the noise won't all be in one direction.

Jon - That's a very helpful GIF to understanding the benefits of drizzle

 

Thank you all - that is exactly the information I was looking for. 


  • bobzeq25 likes this

#6 ks__observer

ks__observer

    Surveyor 1

  • *****
  • Posts: 1,819
  • Joined: 28 Sep 2016
  • Loc: Long Island, New York

Posted 26 October 2021 - 10:39 PM

I image with a SV70 at f4, 3.4 arc-sec/pix.

My stars are all squares.

A 2x Drizzle smoothes things out nicely.

See M52 stars in my latest shot:

https://www.astrobin...6/0/?mod=&real=


  • Benschop likes this

#7 jacquesdeacon

jacquesdeacon

    Sputnik

  • -----
  • Posts: 43
  • Joined: 09 Aug 2011
  • Loc: Kathu, NC, South Africa

Posted 27 October 2021 - 01:14 AM

Cool discussion,
 
Love Ron Rista’s GIF showing those stars tighten up with the combination of deconvolution. 
 
I personally do dither, but have never attempted drizzle.
 
From experience what stack size would you recommend at minimum to see good results with drizzle? I know especially the CMOS + fast optics imagers run huge sub 1 min stacks through this, but not so feasible for us higher read-noise ccd imagers.

 

Regards,

Jacques


Edited by jacquesdeacon, 27 October 2021 - 01:16 AM.

  • Benschop likes this

#8 Benschop

Benschop

    Viking 1

  • *****
  • topic starter
  • Posts: 508
  • Joined: 12 Jan 2015
  • Loc: Ontario

Posted 27 October 2021 - 06:09 PM

That is really helpful to see an example with similar "/pixel. Nice image too.

 

 

I image with a SV70 at f4, 3.4 arc-sec/pix.

My stars are all squares.

A 2x Drizzle smoothes things out nicely.

See M52 stars in my latest shot:

https://www.astrobin...6/0/?mod=&real=


  • ks__observer likes this

#9 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 26,271
  • Joined: 27 Oct 2014

Posted 27 October 2021 - 06:43 PM

Cool discussion,
 
Love Ron Rista’s GIF showing those stars tighten up with the combination of deconvolution. 
 
I personally do dither, but have never attempted drizzle.
 
From experience what stack size would you recommend at minimum to see good results with drizzle? I know especially the CMOS + fast optics imagers run huge sub 1 min stacks through this, but not so feasible for us higher read-noise ccd imagers.

 

Regards,

Jacques

It's not subexposure that counts to achieve the snr improvement needed for drizzle, it's total imaging time.  How many total photons you gather is the big deal, how you break it into pieces counts for a lot less.

 

People have a hard time wrapping their head around that one, but it's true.

 

So, to get the snr necessary for drizzle, simply shoot your usual subexposures, and 2X the number of subs.


  • jacquesdeacon and Benschop like this

#10 ks__observer

ks__observer

    Surveyor 1

  • *****
  • Posts: 1,819
  • Joined: 28 Sep 2016
  • Loc: Long Island, New York

Posted 28 October 2021 - 05:57 AM

jacquesdeacon, on 27 Oct 2021 - 02:14 AM, said:

From experience what stack size would you recommend at minimum to see good results with drizzle? I know especially the CMOS + fast optics imagers run huge sub 1 min stacks through this, but not so feasible for us higher read-noise ccd imagers.

Regards,
Jacques

Good question.
Just did a quick Google on subject but did not see anything useful.
It seems according to one person that about 10 dithers is what you want to shoot for -- about the same as what you want for clearing FPN with dithers.


Edited by ks__observer, 28 October 2021 - 05:57 AM.

  • jacquesdeacon likes this

#11 ks__observer

ks__observer

    Surveyor 1

  • *****
  • Posts: 1,819
  • Joined: 28 Sep 2016
  • Loc: Long Island, New York

Posted 28 October 2021 - 08:32 AM

That is really helpful to see an example with similar "/pixel. Nice image too.

Before and after below:

 

Before:

gallery_263201_9741_68497.jpg

 

After:

gallery_263201_9741_152727.jpg


  • Benschop likes this

#12 jacquesdeacon

jacquesdeacon

    Sputnik

  • -----
  • Posts: 43
  • Joined: 09 Aug 2011
  • Loc: Kathu, NC, South Africa

Posted 28 October 2021 - 02:11 PM

 

 

It's not subexposure that counts to achieve the snr improvement needed for drizzle, it's total imaging time.  How many total photons you gather is the big deal, how you break it into pieces counts for a lot less.

People have a hard time wrapping their head around that one, but it's true.

So, to get the snr necessary for drizzle, simply shoot your usual subexposures, and 2X the number of subs.

Thanks Bob,

 

I am one of them smile.gif

 

For the drizzling black magic to work the algorithm needs dithered (slightly ‘offset’) sub frames to compute the interpolation (not really interpolation - but bear with me) to turn square stars into round stars.

 

So to my logic the algorithm will perform much better with 60 1min exposures than 4 15min exposures (both 60min total exposure) because it has so much more ‘offset information’ to work with.

 

Do you guys agree with this simplified logic or am I missing some fundamental concept?

 

Great discussion!


Edited by jacquesdeacon, 28 October 2021 - 02:11 PM.

  • Benschop likes this

#13 bobzeq25

bobzeq25

    ISS

  • *****
  • Posts: 26,271
  • Joined: 27 Oct 2014

Posted 28 October 2021 - 04:23 PM

Thanks Bob,

 

I am one of them smile.gif

 

For the drizzling black magic to work the algorithm needs dithered (slightly ‘offset’) sub frames to compute the interpolation (not really interpolation - but bear with me) to turn square stars into round stars.

 

So to my logic the algorithm will perform much better with 60 1min exposures than 4 15min exposures (both 60min total exposure) because it has so much more ‘offset information’ to work with.

 

Do you guys agree with this simplified logic or am I missing some fundamental concept?

 

Great discussion!

It, like so much of AP, is complex, with many factors at work.

 

4X15' does not provide for effective dithering, and will likely saturate a lot of stars, ruining star color.

 

60X60" might work.  Or it might give you too much read noise which accumulates per sub.  It's more likely to be correct in worse light pollution and with faster optics. 

 

This _is_ rocket science.  <smile>


  • jacquesdeacon and Benschop like this

#14 ntph

ntph

    Messenger

  • -----
  • Posts: 432
  • Joined: 20 Oct 2012
  • Loc: Sudbury, ON

Posted 28 October 2021 - 11:38 PM

I am under the impression that drizzle doesn't really help things much if image scale is below, say, 2 arc seconds per pixel. I have never tried it; my system works out about 1.2 as/px, so taking the gospel as truth, I have avoided it. Have I interpreted the gospel correctly, or should I be drinking the Drizzle Kool-Aid? Seeing the fairly dramatic differences Jon shows in his example, it is very tempting to give it a try. I am curious though what image scale this is working with? 



#15 jacquesdeacon

jacquesdeacon

    Sputnik

  • -----
  • Posts: 43
  • Joined: 09 Aug 2011
  • Loc: Kathu, NC, South Africa

Posted 29 October 2021 - 01:49 AM

Thanks Bob,

 

 

This _is_ rocket science.  <smile>

This is definitely true! Also, everything works together in the end, drizzling for up-sampling, dithering for outlier rejection, and stacking for SNR improvement, cant really look at these in separate little ‘boxes’.

 

 

I am under the impression that drizzle doesn't really help things much if image scale is below, say, 2 arc seconds per pixel. I have never tried it; my system works out about 1.2 as/px, so taking the gospel as truth, I have avoided it. Have I interpreted the gospel correctly, or should I be drinking the Drizzle Kool-Aid? Seeing the fairly dramatic differences Jon shows in his example, it is very tempting to give it a try. I am curious though what image scale this is working with?

Norm, yes that is very true. Drizzling is best applied to under-sampled data.  You can still apply it to well or over-sampled data, but you wont see that much of a effect. The easiest way to determine this is to look at your data. If your small stars look very blocky, then you are undersampled and drizzling has the potential to recover some of the ‘lost’ resolution.

 

For example, my system’s unbinned resolution is 2.1as/px, reasonably sampled for my seeing. Binned 2x2 my resolution is now 4.2as/px, my stars get blocky as I am undersampling. Drizzling has more potential to recover some details from the latter than the former.


  • Benschop likes this

#16 WadeH237

WadeH237

    Cosmos

  • *****
  • Posts: 7,904
  • Joined: 24 Feb 2007
  • Loc: Ellensburg, WA

Posted 29 October 2021 - 09:22 AM

This _is_ rocket science.  <smile>

Curious.  I have never calculated specific impulse for any of my imaging systems...



#17 Jared

Jared

    Fly Me to the Moon

  • *****
  • Moderators
  • Posts: 7,032
  • Joined: 11 Oct 2005
  • Loc: Piedmont, California, U.S.

Posted 29 October 2021 - 09:44 AM

Curious. I have never calculated specific impulse for any of my imaging systems...


Wow, I sure have. When I get really frustrated with an intransigent piece of equipment and want to boost it into orbit…
  • Jon Rista and Benschop like this


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