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

ASI1600 to bin or nah?

  • Please log in to reply
24 replies to this topic

#1 FiremanDan

FiremanDan

    Aurora

  • *****
  • topic starter
  • Posts: 4732
  • Joined: 11 Apr 2014
  • Loc: Virginia

Posted 30 September 2016 - 10:26 AM

So, Jon's sanity check post reminded me that I have had issues getting strong enough SII and OIII. I know with a normal CCD, binning can be done to basically increase sensitivity, and if I understand correctly it is sometimes at the detriment to resolution/sharpness. I am already at 2.04" pixel scale (not sure what my seeing is or how to measure it, pretty sure it's not good) and I am concerned that my resolution will suffer, but I am also confused as there was a lot of discussion on the software binning on the 1600 vs hardware binning with a "normal" CCD.

My most recent NGC7000 I did a starless tonemap with my NB data (Ha and Oiii) and then used did a LRGB combine with the Ha processed as Lum. But Jon's last post reminded me that different filters show different structures.

 

So in short, how would you decide to bin or not.

Is the ASI1600 a good camera to bin with?



#2 nmoushon

nmoushon

    Apollo

  • *****
  • Moderators
  • Posts: 1363
  • Joined: 12 Oct 2011
  • Loc: Wenatchee Valley, WA

Posted 30 September 2016 - 10:54 AM

I asked this question in an early post when trying to decide on a camera. Find my thread here. Scroll down a couple posts and I bring it up. Several people have some good answers. 


  • FiremanDan likes this

#3 bobzeq25

bobzeq25

    Hubble

  • *****
  • Posts: 16063
  • Joined: 27 Oct 2014

Posted 30 September 2016 - 11:16 AM

I'm going to repeat my answer for convenience.

 

I don't believe the 1600 offers hardware binning.

 

You can bin it in software.  That provides the signal advantage of 2X larger pixels, but not the read noise advantage of a CCD binned in hardware.

 

But, the 1600 has low read noise, so that's less important.

 

The only software binning I'm familiar with is Startools.  StarTools offers variable software binning, and they claim they have a more sophisticated method than most.

 

http://www.startools.org/modules/bin

 

Since PixInsight does everything <grin>, presumably it can bin.


Edited by bobzeq25, 30 September 2016 - 11:18 AM.


#4 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 11:53 AM

The "hardware" binning on the ASI1600 is actually done post-ADC. It is done in hardware, however from an SNR standpoint there is no benefit. Not that pre-ADC binning in charge or voltage domain would really lessen read noise all that much...it is already so low, the reduction with true hardware binning would not seem as great as with a camera that had 9e- read noise.

However, there is one thing about hardware binning with the ASI1600...it forces 10-bit readout. This is not something ZWO chose to do...it is a trait of the sensor itself. While hardware binning can certainly let you get faster frame rates (up to 400fps with smaller ROI), the loss in bit depth can cost you fine details, as you end up getting even greater quantization. The driver supports software binning, which is just the same as post-process downsampling, with the small added benefit that your original files will be smaller, use less space, and will pre-process faster.

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.
  • robedude likes this

#5 nmoushon

nmoushon

    Apollo

  • *****
  • Moderators
  • Posts: 1363
  • Joined: 12 Oct 2011
  • Loc: Wenatchee Valley, WA

Posted 30 September 2016 - 12:12 PM

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.

I would add though that it "can" help when trying to get a better pixel ratio. i.e. imaging with a 9.25" SCT and small galaxies. 



#6 FiremanDan

FiremanDan

    Aurora

  • *****
  • topic starter
  • Posts: 4732
  • Joined: 11 Apr 2014
  • Loc: Virginia

Posted 30 September 2016 - 12:54 PM

Good to know, minimal benefit.
It might actually be clear tonight so maybe, I can do some
work tonight!
I think I'm going to try some high gain work.

#7 FiremanDan

FiremanDan

    Aurora

  • *****
  • topic starter
  • Posts: 4732
  • Joined: 11 Apr 2014
  • Loc: Virginia

Posted 30 September 2016 - 12:56 PM

Disregard, the forecast changed in just the last hour. Went from mostly clear to rain.

#8 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 01:08 PM

 

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.

I would add though that it "can" help when trying to get a better pixel ratio. i.e. imaging with a 9.25" SCT and small galaxies. 

 

I don't know if that is necessarily true if you aren't also getting an SNR benefit. Binning is really just redistributing information. It is trading spatial resolution for SNR. However, if you are not binning in charge or voltage domain where you could theoretically gain a benefit in lowering read noise, then binning in the driver or in the camera is not any different than downsampling in post. The SNR benefit is the same either way. 

 

However, if you downsample, you actually have the opportunity to first apply noise reduction to the original data BEFORE you "bin"...that, as it so happens, would be much like charge or voltage binning. If you have say 10e- total noise in each pixel, and you apply NR that reduces that by a factor of three. You then end up with 3.33e- noise in each pixel. Then you downsample by a factor of 2x, which is basically the same as binning 2x2. If you do not perform NR first, your SNR improvement is 2x. If you do perform NR first and reduce the original noise by a factor of three, then your SNR after downsampling improves by a total factor of 3.45x instead.


Edited by Jon Rista, 30 September 2016 - 01:13 PM.

  • nmoushon likes this

#9 Thirteen

Thirteen

    Gemini

  • *****
  • Posts: 3156
  • Joined: 12 Jul 2013
  • Loc: Milford, Michigan

Posted 30 September 2016 - 01:32 PM

I'll second what Jon says. I use the camera in a somewhat oversampled setup of just over .5"/px. In my experience, It's better to capture and process at the full resolution. This tends to help improve the result when doing both noise reduction and deconvolution. I typically downsample as my last step in the process, thus picking up the additional (and often striking) noise reduction benefit at that time.
  • Tsarbomba and nmoushon like this

#10 FiremanDan

FiremanDan

    Aurora

  • *****
  • topic starter
  • Posts: 4732
  • Joined: 11 Apr 2014
  • Loc: Virginia

Posted 30 September 2016 - 02:02 PM

Downsamplling? Why would you want to lower the sampling. Won't that lower the resolution and lose detail. Do you do any NR, or just downsample?
How would you downsample? Is that done in PI?

#11 Thirteen

Thirteen

    Gemini

  • *****
  • Posts: 3156
  • Joined: 12 Jul 2013
  • Loc: Milford, Michigan

Posted 30 September 2016 - 03:07 PM

Well, yes, it lowers the resolution, but this is in response to this:

 

 

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.

I would add though that it "can" help when trying to get a better pixel ratio. i.e. imaging with a 9.25" SCT and small galaxies. 

 

 

 

Because I shoot at a small pixel scale of .5"/px, I often see in my final images that there just isn't much visual difference between .5"/px and 1"/px.  This can be for a variety of reasons, but mainly it is related to seeing conditions on the night you imaged.    So if that is the case, why not just downsample it to 1"/px for some nice noise reduction and call it a day?  



#12 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 03:12 PM

Downsamplling? Why would you want to lower the sampling. Won't that lower the resolution and lose detail. Do you do any NR, or just downsample?
How would you downsample? Is that done in PI?

Reduce the size of the image. Downscale. Resize the image to make it smaller.



#13 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 03:15 PM

Well, yes, it lowers the resolution, but this is in response to this:

 

 

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.

I would add though that it "can" help when trying to get a better pixel ratio. i.e. imaging with a 9.25" SCT and small galaxies. 

 

 

 

Because I shoot at a small pixel scale of .5"/px, I often see in my final images that there just isn't much visual difference between .5"/px and 1"/px.  This can be for a variety of reasons, but mainly it is related to seeing conditions on the night you imaged.    So if that is the case, why not just downsample it to 1"/px for some nice noise reduction and call it a day?  

If you actually measured the FWHMs, I would be willing to bet that the 0.5" image still has smaller FWHMs despite the seeing. ;) If your seeing is particularly bad, the improvement might be marginal...but my guess is there would still be an improvement. If you deconvolved the higher resolution image first, then downsampled, that could give you even better detail in the resampled image as well. 

 

Lot of reasons to image at 1x1 and handle the "binning" in post. 



#14 bobzeq25

bobzeq25

    Hubble

  • *****
  • Posts: 16063
  • Joined: 27 Oct 2014

Posted 30 September 2016 - 04:05 PM

Downsamplling? Why would you want to lower the sampling. Won't that lower the resolution and lose detail. Do you do any NR, or just downsample?
How would you downsample? Is that done in PI?

You can downsample (and gain some noise reduction), if you're oversampled for:

 

the seeing

the optics

the tracking

 

Increasing sampling cannot increase resolution if something else is impacting resolution.  It's much like using "digital zoom" on a terrestrial camera.  An overly magnified image just magnifies blur from other sources.

 

Read the Startools webpage I cited above.  They show an actual two image comparison showing how, in certain circumstances, you can trade "resolution" for a better to signal to noise ratio.  It's serious proof.  I've used their Binning tool (a form of downsampling) successfully.  The gain was modest, but noticeable.

 

Is it an ideal case to make their point?  No doubt.  But is it a realistic case?  Also, no doubt.


Edited by bobzeq25, 30 September 2016 - 04:07 PM.


#15 NyxAither

NyxAither

    Apollo

  • *****
  • Posts: 1355
  • Joined: 11 Sep 2007

Posted 30 September 2016 - 04:56 PM

Here's how to do software binning in PI: https://pixinsight.c...erResample.html, and if you downscale your image by using the "average" mode in IntegerResample, then you will get a smaller image with improved SNR. It's not as much of an improvement as pre-ADC hardware binning (that the ASI1600 doesn't offer), but it helps, and if you're oversampled anyway then why not ¯\_(ツ)_/¯.


Edited by NyxAither, 30 September 2016 - 05:02 PM.

  • FiremanDan and ks__observer like this

#16 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 05:06 PM

 

Downsamplling? Why would you want to lower the sampling. Won't that lower the resolution and lose detail. Do you do any NR, or just downsample?
How would you downsample? Is that done in PI?

You can downsample (and gain some noise reduction), if you're oversampled for:
 
the seeing
the optics
the tracking
 
Increasing sampling cannot increase resolution if something else is impacting resolution.  It's much like using "digital zoom" on a terrestrial camera.  An overly magnified image just magnifies blur from other sources.
 
Read the Startools webpage I cited above.  They show an actual two image comparison showing how, in certain circumstances, you can trade "resolution" for a better to signal to noise ratio.  It's serious proof.  I've used their Binning tool (a form of downsampling) successfully.  The gain was modest, but noticeable.
 
Is it an ideal case to make their point?  No doubt.  But is it a realistic case?  Also, no doubt.

 

 
Actually, this isn't quite true. Increasing sampling can improve resolution, even if seeing is your single worst blur factor. Total blur is an RMS. It isn't just seeing, it isn't just image scale, it isn't just diffraction nor aberrations, it isn't just guiding and tracking error. It is all of them combined:

TotalBlur = SQRT(Seeing^2 + Diffraction^2 + Aberrations^2 + FilterBlur^2 + GuidingError^2 + TrackingError^2 + WindBlur^2 + PixelScale^2)

If you have an 80mm f/6 scope, your dawes limit is 1.44". If you have a 4 micron pixel size, your image scale is 1.72". Let's say your seeing is 2". Let's say you are using no filters, your lens is truly diffraction limited, you have no wind and your guide RMS is 0.8". Your total system blur would then be:

TotalBlur = SQRT(2^2 + 1.44^2 + 1.72^2 + 0.8^2) = 3.1" RMS

So what happens if you reduce your pixel size by a factor of 2? Your image scale is now 0.86":

TotalBlur = SQRT(2^2 + 1.44^2 + 0.86^2 + 0.8^2) = 2.7" RMS

That right there is a 13% increase in resolution. What if your seeing was a little better than you thought it was? Say 1.8"? This is often what you find as you start better sampling your stars...the pixels themselves add their own blur, so your measurements of what your seeing likely is are going to be skewed (and on top of that, few people account for the diffraction blurring or guide RMS when they try to use FWHM to figure out what their seeing is...most just assume that FWHM === seeing! That's false!!)

TotalBlur = SQRT(1.8^2 + 1.44^2 + 0.86^2 + 0.8^2) = 2.6" RMS

Well, that's a 4% increase...but, your gains are limited. By seeing? Well, not really, seeing just got better. Your scope is still limiting you to 1.44". What if you used a larger aperture? Say a 101mm?

TotalBlur = SQRT(1.8^2 + 1.13^2 + 0.68^2 + 0.8^2) = 2.37" RMS

That is another 9% increase in resolution. Relative to the original 3.1", it's a 24% improvement in resolution. Seeing has hardly changed here...you thought it was 2", turned out it was really 1.8" once you were sampling the stars more accurately and getting more accurate measaurements...that is only a difference of 0.2". Guiding didn't change either, since seeing really didn't change, so your guide RMS is still 0.8". What really changed was the sampling ratio...better sampling leads to better resolution most of the time. Given this is an RMS, the larger terms do dominate. If you have PARTICULARLY bad seeing, which I consider to be around 3" or worse (sadly, farily common in more northern latitudes, i.e. northern Canada, Netherlands, Alaska, etc. where you might be stuck under the polar vortex), then seeing will certainly limit your results and diminish the returns on better sampling. However I believe true 3" seeing is pretty rare...3" FWHMs are not...but FWHM is the TotalBlur...which includes many other factors.


  • NyxAither likes this

#17 bobzeq25

bobzeq25

    Hubble

  • *****
  • Posts: 16063
  • Joined: 27 Oct 2014

Posted 30 September 2016 - 05:45 PM

OK, good theory.  What do you have to say about the actual before/after image on the StarTools web page?



#18 anismo

anismo

    Fly Me to the Moon

  • *****
  • Posts: 5930
  • Joined: 30 Jan 2014
  • Loc: Austin, TX, USA

Posted 30 September 2016 - 06:09 PM

Jon..just something that popped up when I scanned your equations. You are missing the "mean" portion of RMS.. so in your case it looks like you are squaring the values and adding them but not dividing them by number of elements before the square root.

 

or perhaps you werent going for RMS.



#19 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 06:14 PM

OK, good theory.  What do you have to say about the actual before/after image on the StarTools web page?

They look fake. Binning trades resolution for SNR. However, those two images have exactly the same pixel dimensions. I am not saying the feature is fake, however the images do not appear to be one high res and the other "binned" from it, as if they were, the high res image would have four times the pixels. 



#20 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 06:20 PM

Jon..just something that popped up when I scanned your equations. You are missing the "mean" portion of RMS.. so in your case it looks like you are squaring the values and adding them but not dividing them by number of elements before the square root.

 

or perhaps you werent going for RMS.

True. I guess it's not actually an RMS (it's a root of squares...? Is there even a name for this?), however I do believe this is the proper formula, as if I do take the mean, then that would imply your FWHMs would be 1.18" in the best case there...not a chance of that, given your largest term is still 1.8". Your best possible FWHM will always be larger than your largest blur term. Depending on the scale of the other terms, it could be quite a bit larger. For example, if all your blur terms were 1.5", then your best possible FWHM would be 3". If only one of your blur terms was 1.5", and the rest were 0.1", your best possible FWHM would be 1.51"...it'll always be larger than your largest blur term. (In other words, it's an asymptotic relationship, FWHM to largest blur term).


Edited by Jon Rista, 30 September 2016 - 06:26 PM.


#21 anismo

anismo

    Fly Me to the Moon

  • *****
  • Posts: 5930
  • Joined: 30 Jan 2014
  • Loc: Austin, TX, USA

Posted 30 September 2016 - 06:44 PM

Ya that is what I thought. It probably has a different name. So just didn't want anyone tripping over the term RMS.



#22 Thirteen

Thirteen

    Gemini

  • *****
  • Posts: 3156
  • Joined: 12 Jul 2013
  • Loc: Milford, Michigan

Posted 30 September 2016 - 07:15 PM

 

Well, yes, it lowers the resolution, but this is in response to this:

 

 

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.

I would add though that it "can" help when trying to get a better pixel ratio. i.e. imaging with a 9.25" SCT and small galaxies. 

 

 

 

Because I shoot at a small pixel scale of .5"/px, I often see in my final images that there just isn't much visual difference between .5"/px and 1"/px.  This can be for a variety of reasons, but mainly it is related to seeing conditions on the night you imaged.    So if that is the case, why not just downsample it to 1"/px for some nice noise reduction and call it a day?  

If you actually measured the FWHMs, I would be willing to bet that the 0.5" image still has smaller FWHMs despite the seeing. ;) If your seeing is particularly bad, the improvement might be marginal...but my guess is there would still be an improvement. If you deconvolved the higher resolution image first, then downsampled, that could give you even better detail in the resampled image as well. 

 

Lot of reasons to image at 1x1 and handle the "binning" in post. 

Sorry, not sure how it came across.  I'm in agreement with you that generally the .5" will probably have slightly lower FWHM and handle deconvolution better.   I'm saying that at the end of my processing, I will inspect the full size image and also another version at 50% or something close to that.  I'll generally reduce the image size if it doesn't hurt the aesthetics in a noticeable way and post that to the web.  

 

I'll save the full size off for printing ;)



#23 Jon Rista

Jon Rista

    ISS

  • *****
  • Posts: 23289
  • Joined: 10 Jan 2014
  • Loc: Colorado

Posted 30 September 2016 - 07:33 PM

 

 

Well, yes, it lowers the resolution, but this is in response to this:

 

 

So, generally speaking, there is not a lot of real-world benefit to using binning while you are imaging. The biggest benefit for DSO imaging might be less disk space used and faster preprocessing.

I would add though that it "can" help when trying to get a better pixel ratio. i.e. imaging with a 9.25" SCT and small galaxies. 

 

 

 

Because I shoot at a small pixel scale of .5"/px, I often see in my final images that there just isn't much visual difference between .5"/px and 1"/px.  This can be for a variety of reasons, but mainly it is related to seeing conditions on the night you imaged.    So if that is the case, why not just downsample it to 1"/px for some nice noise reduction and call it a day?  

If you actually measured the FWHMs, I would be willing to bet that the 0.5" image still has smaller FWHMs despite the seeing. ;) If your seeing is particularly bad, the improvement might be marginal...but my guess is there would still be an improvement. If you deconvolved the higher resolution image first, then downsampled, that could give you even better detail in the resampled image as well. 

 

Lot of reasons to image at 1x1 and handle the "binning" in post. 

Sorry, not sure how it came across.  I'm in agreement with you that generally the .5" will probably have slightly lower FWHM and handle deconvolution better.   I'm saying that at the end of my processing, I will inspect the full size image and also another version at 50% or something close to that.  I'll generally reduce the image size if it doesn't hurt the aesthetics in a noticeable way and post that to the web.  

 

I'll save the full size off for printing ;)

 

Ah, gocha! Yeah, I often do the same, resample at the end of my processing. Sometimes the full size image just doesn't hold up, and yet a 2x downsample looks great. 



#24 dts350z

dts350z

    Sputnik

  • *****
  • Posts: 38
  • Joined: 31 Mar 2014

Posted 18 March 2018 - 11:09 PM

Well this is an "old" thread but I read all the asi1600 stuff with great interest and currently have this question, based on this thread.

 

I can see that the star size is smaller, measured seeing (AstroImageJ) is better, when binning 1x1 vs. 2x2. I can see that there is more dynamic range as well.

 

However, the stdDev of any piece of "sky" (no stars or nebulosity) is greater at 1x1. Even if I divide the StdDev of the "sky" in the 1x1 image by 4 (12 bits vs. 10 bits in the 2x2 image) I still have a higher value in the 1x1.

 

Note I also had to increase the gain from 200 to 300 to keep to 10 min exposure with ZWO ha filter. One other difference is the moon was up during the 2x2, but not during the 1x1. 

 

So, back to the question of this thread, should I bin or no, given an increase in noise?



#25 rottielover

rottielover

    Vostok 1

  • -----
  • Posts: 114
  • Joined: 11 Jun 2017
  • Loc: St. Louis MO

Posted 24 March 2018 - 04:33 PM

Well this is an "old" thread but I read all the asi1600 stuff with great interest and currently have this question, based on this thread.

 

I can see that the star size is smaller, measured seeing (AstroImageJ) is better, when binning 1x1 vs. 2x2. I can see that there is more dynamic range as well.

 

However, the stdDev of any piece of "sky" (no stars or nebulosity) is greater at 1x1. Even if I divide the StdDev of the "sky" in the 1x1 image by 4 (12 bits vs. 10 bits in the 2x2 image) I still have a higher value in the 1x1.

 

Note I also had to increase the gain from 200 to 300 to keep to 10 min exposure with ZWO ha filter. One other difference is the moon was up during the 2x2, but not during the 1x1. 

 

So, back to the question of this thread, should I bin or no, given an increase in noise?

What I have taken away after reading this and other threads on the topic:

 

In a nutshell, there's only a few circumstances when capturing data where you'd want to bin with these camera's (I have one and have experimented with 2x2 vs 1x1 to a limited degree).  So few that for practical purposes there seems to be no reason to capture on anything but 1x1 and leave it at that.

 

Now that said, I have made some very good images (at least if I say so myself), by capturing Lum or Narrow Band at 1x1 and RGB at 2x2.  But this does increase the work you do in processing (at least in PI) and you need separate dark and bias frames to match.

 

All in all after having the 1600mm-cool for about 6 months,  I'm going to opt to the 1x1, leave it, and have slightly less to worry about and fiddle with  :)

 

Hope that helps.




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