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 Intergration Pros and Cons

  • Please log in to reply
26 replies to this topic

#1 Becomart

Becomart

    Apollo

  • -----
  • topic starter
  • Posts: 1,081
  • Joined: 03 Jan 2015

Posted 23 January 2020 - 06:15 PM

I’ve started to use drizzle intergration in my stacking process. My understanding is that it interpolates the image. I know some do this and then mix back down later. Any pros and cons of this procedure. Anyone deliberately not using drizzle and can explain why?



#2 gezak22

gezak22

    Gemini

  • *****
  • Posts: 3,168
  • Joined: 15 Aug 2004
  • Loc: On far side of moon. Send help.

Posted 23 January 2020 - 06:30 PM

Dither + 2x drizzle always seemed like a no-brainer to me.


  • psandelle and elmiko like this

#3 sharkmelley

sharkmelley

    Mercury-Atlas

  • *****
  • Posts: 2,840
  • Joined: 19 Feb 2013

Posted 23 January 2020 - 06:57 PM

For OSC and DSLR cameras, CFA Drizzle (aka Bayer Drizzle) gives improved resolution because it avoids the interpolation performed during debayering.  It also gives a more finely grained noise profile which is more pleasant to the eye than the "lumpy" noise from debayering.

 

Mark


  • psandelle, elmiko, Jon Rista and 1 other like this

#4 Madratter

Madratter

    Voyager 1

  • *****
  • Posts: 11,732
  • Joined: 14 Jan 2013

Posted 24 January 2020 - 01:55 PM

Drizzle makes sense when you are under sampled. I'm not with my AT8RC. I don't. My SV80 could benefit from it marginally.



#5 jdk

jdk

    Ranger 4

  • *****
  • Posts: 328
  • Joined: 06 Jul 2017

Posted 24 January 2020 - 01:58 PM

This is probably a good time to ask about a general threshold for undersampling...any broad-stroke guidelines? Anything over 2"/px? 1.75? 1.5?



#6 Madratter

Madratter

    Voyager 1

  • *****
  • Posts: 11,732
  • Joined: 14 Jan 2013

Posted 24 January 2020 - 02:01 PM



This is probably a good time to ask about a general threshold for undersampling...any broad-stroke guidelines? Anything over 2"/px? 1.75? 1.5?

My rule of thumb on this would be simple. If "too many" of your stars are looking squarish, it would benefit from drizzle. It is up to you how many is "too many".

 

Here is a real life example: (Warning the link is intentionally 100% fullsize).

 

get.jpg?insecure

 

Image scale is 2.288. That is undersampled for my skies.

 

There are some square stars. To me, they are not "too many". YMMV.


Edited by Madratter, 24 January 2020 - 02:08 PM.

  • elmiko, TrustyChords and klaussius like this

#7 OhmEye

OhmEye

    Ranger 4

  • -----
  • Posts: 343
  • Joined: 15 Sep 2019
  • Loc: Western NY Southern Tier

Posted 24 January 2020 - 03:05 PM

Regardless of sampling, although drizzle adds noise, I find it to be finer grain noise that I am able to reduce with less NR and less obvious artifacts. I'm usually more pleased with my results and often use at least 1.5x drizzle on good data just for the noise processing results alone.
  • psandelle, elmiko, sharkmelley and 1 other like this

#8 Fernando134

Fernando134

    Ranger 4

  • -----
  • Posts: 356
  • Joined: 02 Feb 2013
  • Loc: Brasil

Posted 27 January 2020 - 08:47 AM

On Warren A Keller book, "Inside Pixinsight", pag 312: Using the script "FWHMEccentricity"  determine the Median FWHM: a "value under 2 pixels indicates that the image is undersampled, and is a candidate for Drizzle integration".

As I am not an expert, that is what I do with my images in Pixinsight.

 

Hope this help. Regards, Fernando.



#9 spokeshave

spokeshave

    Vanguard

  • *****
  • Posts: 2,169
  • Joined: 08 Apr 2015

Posted 27 January 2020 - 09:05 AM

I’ve started to use drizzle intergration in my stacking process. My understanding is that it interpolates the image. I know some do this and then mix back down later. Any pros and cons of this procedure. Anyone deliberately not using drizzle and can explain why?

One of the valuable features of drizzle is that it is not an interpolation process. It recovers genuine spatial detail that is lost to undersampling by exploiting the additional spatial sampling that dithering allows.

 

Tim


  • Jon Rista and Der_Pit like this

#10 klaussius

klaussius

    Mariner 2

  • *****
  • Posts: 280
  • Joined: 02 Apr 2019
  • Loc: Buenos Aires

Posted 27 January 2020 - 09:15 AM

One of the valuable features of drizzle is that it is not an interpolation process. It recovers genuine spatial detail that is lost to undersampling by exploiting the additional spatial sampling that dithering allows.

 

Tim

It does interpolate. Perhaps you're thinking about interleaving, which doesn't interpolate. Interleaving is hard to do, it depends on very careful dithering. Drizzle overcomes that by interpolating. It can still recover increased spatial detail, though, and that's the beauty of it. It just works.


  • psandelle likes this

#11 dhaval

dhaval

    Vendor

  • *****
  • Vendors
  • Posts: 1,836
  • Joined: 21 Jul 2008
  • Loc: Round Rock, TX

Posted 27 January 2020 - 09:56 AM

I always run PSFImage script in PI to see what the star looks like - if I think it is "squarish" (and hence under-sampled) I will run drizzle on the data. It has worked quite OK till now.

CS!


  • psandelle likes this

#12 Peregrinatum

Peregrinatum

    Viking 1

  • *****
  • Posts: 849
  • Joined: 27 Dec 2018
  • Loc: South Central Valley, Ca

Posted 27 January 2020 - 10:05 AM

dither or die!


  • elmiko likes this

#13 Der_Pit

Der_Pit

    Apollo

  • *****
  • Posts: 1,190
  • Joined: 07 Jul 2018
  • Loc: La Palma

Posted 27 January 2020 - 10:29 AM

It does interpolate. Perhaps you're thinking about interleaving, which doesn't interpolate. Interleaving is hard to do, it depends on very careful dithering. Drizzle overcomes that by interpolating. It can still recover increased spatial detail, though, and that's the beauty of it. It just works.

Depends what you call interpolation.  It projects the source pixel into the (enlarged) target plane and spreads its counts over covered pixels.  But it does not interpolate any source pixel values, I guess that is what Tim referred to.



#14 klaussius

klaussius

    Mariner 2

  • *****
  • Posts: 280
  • Joined: 02 Apr 2019
  • Loc: Buenos Aires

Posted 27 January 2020 - 11:08 AM

Depends what you call interpolation.  It projects the source pixel into the (enlarged) target plane and spreads its counts over covered pixels.  But it does not interpolate any source pixel values, I guess that is what Tim referred to.

I'm using this as my reference. What I'm calling interleaving that page calls interlacing, but the idea is that interlacing is "like drizzle" without the interpolation.

 

Interpolation in the sense that each image pixel lands on more than 1 target pixel, with different weights, the result of which is pretty much like interpolating the smaller image into the bigger dimensions. In interlacing, each image pixel lands on 1 target pixel only.


Edited by klaussius, 27 January 2020 - 11:08 AM.


#15 Der_Pit

Der_Pit

    Apollo

  • *****
  • Posts: 1,190
  • Joined: 07 Jul 2018
  • Loc: La Palma

Posted 27 January 2020 - 03:51 PM

Yes, but this is not interpolation.  Interpolation is done in normal alignment, where you look up where in the raw image the information for a pixel in the aligned image is coming from.   The value at that position (which usually is fractional in pixels) is then interpolated from surrounding source pixels.  So you use backward projection, and interpolation of the raw data.

Drizzle uses forward projection from raw image space to aligned space, and there just (geometrically) splits the raw intensity in various pieces.  But no interpolation, and therefore it is not endangered by halo/ringing effects, like interpolation is.  

Which is probably the reason why - as stated here by others - the background looks nicer for drizzle: because interpolating noisy data is cumbersome.

 

Your interleaving (all of a source pixel goes to one target pixel) is indeed drizzling with a reduction factor of zero (the 'raindrop' has no extent and always hits only one pixel)



#16 gregwjones

gregwjones

    Sputnik

  • -----
  • Posts: 28
  • Joined: 07 Feb 2018
  • Loc: Texas

Posted 27 January 2020 - 04:12 PM

I am currently taking a fresh look at my dithering approach and settings. I know that drizzle integration requires the set of subframes to be dithered.

 

But, what size of dither and how many different dither positions does the drizzle integration process require or prefer?

 

Is a simple spiral pattern better for drizzle or does it work better with a random pattern?

 

Regards,

Greg



#17 klaussius

klaussius

    Mariner 2

  • *****
  • Posts: 280
  • Joined: 02 Apr 2019
  • Loc: Buenos Aires

Posted 27 January 2020 - 04:29 PM

I am currently taking a fresh look at my dithering approach and settings. I know that drizzle integration requires the set of subframes to be dithered.

 

But, what size of dither and how many different dither positions does the drizzle integration process require or prefer?

 

Is a simple spiral pattern better for drizzle or does it work better with a random pattern?

 

Regards,

Greg

If you dither every frame (or every few frames), random works just fine.

 

If you don't dither much, the pattern starts to matter a bit more. But it's very difficult to get good results with insufficient dithering, so don't hold back on it.

 

Drizzle doesn't need more than 1 pixel dither, so dither distance will depend more on your camera's FPN than drizzle.

 

Drizzle xN requires N² dither positions at least, so make sure you dither way more than that minimum. Ie: if you're taking 5min subs, and dither every 5 subs, and capture 2h of data, it's unlikely you'll be able to drizzle, as you would have only 5 dither positions, and that's unlikely to work very well with drizzle 2x. If you dither every frame, you'll be fine.


Edited by klaussius, 27 January 2020 - 04:31 PM.


#18 Biggen

Biggen

    Ranger 4

  • *****
  • Posts: 350
  • Joined: 04 Jan 2019

Posted 27 January 2020 - 07:16 PM

I’m still confused on when to drizzle. If I’m imaging at 2.6 sec/pixel is that considered under sampled enough to warrant drizzling?


Edited by Biggen, 27 January 2020 - 07:16 PM.

  • elmiko likes this

#19 elmiko

elmiko

    Skylab

  • *****
  • Posts: 4,064
  • Joined: 27 Jul 2009
  • Loc: Arizona

Posted 27 January 2020 - 07:41 PM

I’m still confused on when to drizzle. If I’m imaging at 2.6 sec/pixel is that considered under sampled enough to warrant drizzling?

Just zoom in on one of your images. If you're stars show alot of pixelation...blocky looking then Drizzle integration will definitely improve your images! I just started drizzling about 8 months ago, I would have done it from the beginning if I would have known it would be so easy to do! 


  • Biggen and klaussius like this

#20 Biggen

Biggen

    Ranger 4

  • *****
  • Posts: 350
  • Joined: 04 Jan 2019

Posted 27 January 2020 - 07:46 PM

Just zoom in on one of your images. If you're stars show alot of pixelation...blocky looking then Drizzle integration will definitely improve your images! I just started drizzling about 8 months ago, I would have done it from the beginning if I would have known it would be so easy to do! 

I use APP for integration so I need to figure out how to drizzle with it. Something else to learn...


  • elmiko likes this

#21 elmiko

elmiko

    Skylab

  • *****
  • Posts: 4,064
  • Joined: 27 Jul 2009
  • Loc: Arizona

Posted 27 January 2020 - 07:53 PM

Are you dithering during aquisition? You have to Dither before you can drizzle integration. I use Pixinsight, don't know about APP but I am sure it can be done. Check the manual.


  • Biggen likes this

#22 james7ca

james7ca

    Cosmos

  • *****
  • Posts: 7,900
  • Joined: 21 May 2011
  • Loc: San Diego, CA

Posted 27 January 2020 - 08:04 PM

I’m still confused on when to drizzle. If I’m imaging at 2.6 sec/pixel is that considered under sampled enough to warrant drizzling?

It isn't the arc seconds per pixel that matters (as any absolute cutoff), it's the number of pixels that span the spot size of your optics (where "spot size" can be taken as the size of the star's Airy disk after the adverse effects of the seeing conditions). How that value relates to the image scale in arc seconds per pixel depends upon the aperture of your scope.

 

That said, many people use a value of two arc seconds as a typical star size given average seeing and thus in that case you'd want to be imaging at around one arc second per pixel (as a common limit for a moderate to large scope). However, a small scope isn't going to be able to produce an Airy disk that small and thus you won't gain much if anything by imaging at that scale. As previously mentioned (and only approximately) you generally want to sample a star with at least two pixels.

 

Here are a few Airy disk sizes in green light at various apertures:

 

50mm: 5.1"

80mm: 3.2"

100mm: 2.6"

150mm: 1.7"

200mm: 1.3"

 

In actual practice, it's somewhat unlikely that you're going to see much improvement in resolution or detail when going much beyond 200mm in aperture (at least not with long-exposure DSO work).


  • elmiko, ks__observer and Biggen like this

#23 OhmEye

OhmEye

    Ranger 4

  • -----
  • Posts: 343
  • Joined: 15 Sep 2019
  • Loc: Western NY Southern Tier

Posted 28 January 2020 - 07:58 AM

Drizzling with APP is flexible, powerful and easy. On the integration tab you select drizzle (or bayer drizzle if you are using OSC) and set the image scale and droplet size you want to drizzle. I scale the droplet inverse to the image scale. For example, for a 2x image scale I use a 0.5 droplet size. For 1.5x I use 0.75, or 3x I use 0.34 droplets. The larger you scale, the larger the resolution and the more time processing takes. You can also change the droplet kernel, which is the shape of the drop, but the default looks best to me.

 

https://www.astropix...r-mono-cameras/


  • elmiko, eshy76 and Biggen like this

#24 spokeshave

spokeshave

    Vanguard

  • *****
  • Posts: 2,169
  • Joined: 08 Apr 2015

Posted 28 January 2020 - 09:03 AM

I'm using this as my reference. What I'm calling interleaving that page calls interlacing, but the idea is that interlacing is "like drizzle" without the interpolation.

 

Interpolation in the sense that each image pixel lands on more than 1 target pixel, with different weights, the result of which is pretty much like interpolating the smaller image into the bigger dimensions. In interlacing, each image pixel lands on 1 target pixel only.

Call it whatever you wish, but it's not interpolation. At no point is a guess made at a value between two other values. Drizzle is a very clever way to exploit the fact that dithering is a form of spatial sampling. The data that drizzle extracts is wholly contained within the dithered data. Nothing is interpolated.  

 

Tim



#25 Biggen

Biggen

    Ranger 4

  • *****
  • Posts: 350
  • Joined: 04 Jan 2019

Posted 28 January 2020 - 09:20 AM

Drizzling with APP is flexible, powerful and easy. On the integration tab you select drizzle (or bayer drizzle if you are using OSC) and set the image scale and droplet size you want to drizzle. I scale the droplet inverse to the image scale. For example, for a 2x image scale I use a 0.5 droplet size. For 1.5x I use 0.75, or 3x I use 0.34 droplets. The larger you scale, the larger the resolution and the more time processing takes. You can also change the droplet kernel, which is the shape of the drop, but the default looks best to me.

 

https://www.astropix...r-mono-cameras/

 

 

Hmm..  So the larger the scale, I need to lower the droplet size which I guess increases resolution at the expense of more noise?

 

I guess I'd need to integrate an image with and without drizzle and see if I can see a difference at all.


Edited by Biggen, 28 January 2020 - 09:23 AM.



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