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

Color CMOS binning - basic questions

  • Please log in to reply
30 replies to this topic

#1 gustavo_sanchez

gustavo_sanchez

    Apollo

  • *****
  • Moderators
  • topic starter
  • Posts: 1,234
  • Joined: 30 Dec 2010
  • Loc: Puerto Rico, US

Posted 07 February 2019 - 08:19 PM

Hello guys, when doing bin2 on a color ZWO camera, how the read noise, full well capacity, and dynamic range get affected? How the color arrangement is preserved?

#2 Pauls72

Pauls72

    Mercury-Atlas

  • *****
  • Posts: 2,818
  • Joined: 28 Oct 2007
  • Loc: LaPorte, IN USA

Posted 07 February 2019 - 08:45 PM

CMOS binning is not like hardware binning on CCD cameras. The vast majority of CMOS cameras only do software binning. So all it does is reduce the data set and file size and it has no effect on read noise, full well capacity, and dynamic range. So if you do 2x2 software binning, the values from 4 pixels are added together and then divided by 4.

 

In CCD hardware binning, the photosites (pixels) are electrically tided together and act as one on the hardware level.


  • OleCuss likes this

#3 OleCuss

OleCuss

    Mercury-Atlas

  • *****
  • Posts: 2,655
  • Joined: 22 Nov 2010

Posted 07 February 2019 - 08:46 PM

Well, I'm sure others will check in but I want to point out a possible minor confusion.

 

The ASI071 is unusual but you can actually do a sort of hardware binning with that IMX071 sensor.  It appears to convey no benefit other than a slightly faster frame rate and I've read that you may get artifacts or more noise.  That hardware binning is not recommended unless you simply must have that faster frame rate.

 

But in general CMOS binning will not be in-camera but will be in software.  This means there will be no decrease in read noise.  There will be no improvement in dynamic range.  I think there may be some decrease in shot noise?

 

This will be generally true for all the CMOS cameras with those unusual exceptions where there is that sort of hardware binning which you'll likely want to actually avoid.



#4 gustavo_sanchez

gustavo_sanchez

    Apollo

  • *****
  • Moderators
  • topic starter
  • Posts: 1,234
  • Joined: 30 Dec 2010
  • Loc: Puerto Rico, US

Posted 07 February 2019 - 09:07 PM

I read there was some advantage to the SNR, albeit no as efficient as the hardware binning of a CCD.

#5 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10,586
  • Joined: 03 Jul 2004

Posted 07 February 2019 - 09:39 PM

TL; DR

This thread is too strong a dose:   https://www.cloudyni...ing-in-my-book/

 

In a nutshell, hardware binning (in my book, using my classic definition) is charge domain binning (thus only one read out step is needed.)

This has always been the CCD method of doing the binning.

 

In recent days, CMOS folks want to "reuse" the term but in a different context (you say tom-EI-to and I say tom-AH-to.)   This confuses people.

The gist is:   "it is just mathematical summing (or averaging) values".   

You can do that in photo-processing software, or in the case of, SONY CMOS image sensor, it is done inside the sensor chip itself  (so the goofy term "hardware binning" is reused.)  <-- it is supposed to be more accurately stated as "doing binning software addition in hardware" 

(Guess people can live with a one-pound can coffee only containing 12 ounces, so why not...)

 

P.S. in most SONY EXMOR cases, the sole goal is to make read-out faster (higher frame rate.)

       It does not provide any benefit at all.  In few EXMOR sensor ICs, I've seen the chip adding four like-color 14-bit pixels together, averaging them, then output a 12-bit result.

 

P.P.S. why this type of so-called hardware binning still retain color information

Underneath, it is really 4-by-4 pixels, mathematically averaged into a 2x2 Bayer (RGGB matrix) quad-pixel output.


  • dmdouglass, OleCuss and gustavo_sanchez like this

#6 Pauls72

Pauls72

    Mercury-Atlas

  • *****
  • Posts: 2,818
  • Joined: 28 Oct 2007
  • Loc: LaPorte, IN USA

Posted 07 February 2019 - 09:46 PM

With true 2x2 hardware binning, all 4 photosites (pixels) are electrically tided together as one. So your well depth increases by 4 times, the area of your pixel increases by 4 times, in theory you capture 4 times as many photons in the same time period and you have lower read noise and all 4 photosites are read as 1. So you only have the signal going through 1 amplifier instead of 4 and you get a lower SNR. On the down side you cut your resolution by a factor of 4.

 

What is common with a mono CCD camera, take your luminance subs unbinned (B&W) to get the image details and then take you R, G and B binned to get the color data. You don't need the high amount of detail in the color to produce good images.

 

If you have a OSC CCD camera that does binning, the color data is lost when you bin it.



#7 gustavo_sanchez

gustavo_sanchez

    Apollo

  • *****
  • Moderators
  • topic starter
  • Posts: 1,234
  • Joined: 30 Dec 2010
  • Loc: Puerto Rico, US

Posted 07 February 2019 - 09:58 PM

With true 2x2 hardware binning, all 4 photosites (pixels) are electrically tided together as one. So your well depth increases by 4 times, the area of your pixel increases by 4 times, in theory you capture 4 times as many photons in the same time period and you have lower read noise and all 4 photosites are read as 1. So you only have the signal going through 1 amplifier instead of 4 and you get a lower SNR. On the down side you cut your resolution by a factor of 4.

What is common with a mono CCD camera, take your luminance subs unbinned (B&W) to get the image details and then take you R, G and B binned to get the color data. You don't need the high amount of detail in the color to produce good images.

If you have a OSC CCD camera that does binning, the color data is lost when you bin it.

Paul, I agree with your statements. I (think) I got CCD hardware binning clear. My questions are directed to both CMOS binning in general and to the special case of color (OSC) CMOS binning. Does color gets destroyed? There's nothing to be gained in addition to reduce oversampling (my actual problem with my ASI178MC)?

#8 gustavo_sanchez

gustavo_sanchez

    Apollo

  • *****
  • Moderators
  • topic starter
  • Posts: 1,234
  • Joined: 30 Dec 2010
  • Loc: Puerto Rico, US

Posted 07 February 2019 - 10:04 PM



P.S. in most SONY EXMOR cases, the sole goal is to make read-out faster (higher frame rate.)
It does not provide any benefit at all. In few EXMOR sensor ICs, I've seen the chip adding four like-color 14-bit pixels together, averaging them, then output a 12-bit result.


Can this be elaborated upon a bit more?

#9 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10,586
  • Joined: 03 Jul 2004

Posted 07 February 2019 - 10:36 PM

If  four 14-bit pixels' values are supposed to be averaged out (V1 + V2 + V3 + v4) / 4

truncate out the result's decimal value, it will still be a 14-bit value.

** If you don't do averaging, the result will be a 16-bit value, even better, no truncation.

 

Many EXMOR not only not giving out the 14-bit value, it futher gets ride of the last two LSB to make it as a 12-bit result.

The reason is simple: highest possible frame rate.

 

One might ask, if I use (plug in your most favorite brand of CMOS imager's name here) device driver, I got a 16-bit value.  What gives?

Two possibilities:

1) in the device driver level, it reads all pixels values as-is, for all pixels.  Then inside the same device driver, it performs the summing (or averaging) operation

or

2) the device driver set the EXMOR is such so-called "hardware binning" mode, fetch the resulted 12-bit binned pixel values, then left-shift to make it as 16-bit pixel value

 

** Note that many astroimager's device driver always provide the 16-bit value, no matter what, just to make it nicer.

 

In the case of (1), you'll find there is no numerical gaps in resulted image.  Also, even in short exposure time, the frame rate still is very low.

In the case of (2), check the histogram, ugly toothcomb pattern shows.

 

BTW, why SONY likes 12-bit?  Because it is easy to pack 2 pixels' value into a 3-byte word format.


  • OleCuss likes this

#10 WadeH237

WadeH237

    Fly Me to the Moon

  • *****
  • Posts: 5,747
  • Joined: 24 Feb 2007
  • Loc: Snohomish, WA

Posted 26 March 2019 - 03:45 PM

So if you do 2x2 software binning, the values from 4 pixels are added together and then divided by 4.

So this thread was referenced from a newer thread on CMOS binning.

 

Based on this statement, here is my question:  Is it possible that, in the case of the ASI1600, it's bit shifting each digitized pixel value 2 bits (the divide by 4 part) and then summing them?  If so, is this why the camera says that it is only 10 bits when binned 2x2?



#11 OldManSky

OldManSky

    Soyuz

  • *****
  • Posts: 3,900
  • Joined: 03 Jan 2019
  • Loc: Valley Center, CA USA

Posted 26 March 2019 - 06:48 PM

So this thread was referenced from a newer thread on CMOS binning.

 

Based on this statement, here is my question:  Is it possible that, in the case of the ASI1600, it's bit shifting each digitized pixel value 2 bits (the divide by 4 part) and then summing them?  If so, is this why the camera says that it is only 10 bits when binned 2x2?

Yes, it's possible.

But I suspect it's the other way around (less operations -- and the whole "software binning" thing is for speed):  sum them, then shift the sum by 2 bits. :)



#12 WadeH237

WadeH237

    Fly Me to the Moon

  • *****
  • Posts: 5,747
  • Joined: 24 Feb 2007
  • Loc: Snohomish, WA

Posted 27 March 2019 - 05:48 AM

Yes, it's possible.

But I suspect it's the other way around (less operations -- and the whole "software binning" thing is for speed):  sum them, then shift the sum by 2 bits. smile.gif

You are probably right about that.

 

It's too bad that the driver doesn't sum the 4 x 12 bit pixels into a 16 bit value, since it returns a 16 bit value to the imaging software anyway.  That would make driver binning genuinely useful.  You could have the benefit of reduced file sizes without clipping the data to 10 bits.


  • Jon Rista likes this

#13 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10,586
  • Joined: 03 Jul 2004

Posted 27 March 2019 - 06:14 AM

Some activities can happen inside the SoC imager sensor and some can happen in device driver (software runs in computer) level.

Your question about ASI1600's "2x2 bin output at 10bit" setting and the underlying design (how it is accomplished) will have to be answered by ZWO folks (ask question in QA or its forum.)

 

Note that this thread is mostly about what could happen in the CMOS imager sensor SoC level.   And it;s all about the raw data straight out from the SoC.

A device driver basically is just a software component, written by the astroimager manufacturer at its own discretion (with secret sauce and what's not.)  <-- it is beyond scope of this thread



#14 rzgp33

rzgp33

    Sputnik

  • -----
  • Posts: 34
  • Joined: 05 Sep 2019
  • Loc: Bendigo, Victoria, Australia

Posted 30 October 2019 - 10:02 PM

I know that this is an old thread ... but for those looking for info try here for a fairly authoritative  document.

 

https://www.atik-cam...n-cmos-and-ccd/

 

Regards

Brian



#15 Narrowbandpaul

Narrowbandpaul

    Mariner 2

  • -----
  • Posts: 211
  • Joined: 24 Dec 2013
  • Loc: Constrained by the Heisenberg uncertainty principle

Posted 31 October 2019 - 07:09 PM

I am not sure I would regard any ATIK document as an ‘authoritative’ reference on CMOS sensors. Much more appropriate would be anything from the father of CMOS imagers, Eric Fossum


Edited by Narrowbandpaul, 31 October 2019 - 07:11 PM.


#16 dnick

dnick

    Lift Off

  • -----
  • Posts: 8
  • Joined: 15 Mar 2018
  • Loc: Tewkesbury, UK

Posted 06 November 2019 - 05:36 PM

I know this is a post revived by two recent additions but if anyone is interested Robin on the SharpCap forum wrote a few posts about this subject in July 2017 that are very informative and they show the difference between CCD and CMOS binning and the effect this has on the numbers for both types of sensor.

 

 https://forums.sharp...topic.php?t=262

 

Cheers

Nick


  • donstim likes this

#17 Coconuts

Coconuts

    Viking 1

  • *****
  • Posts: 539
  • Joined: 23 Sep 2012

Posted 06 November 2019 - 08:40 PM

Binning isn't just about read noise.  As long as you are still oversampling properly (so slanted toward longer focal lengths), doubling the pixel size quadruples the etendue.  That is pretty far from "It doesn't provide any benefit at all".  Etendue is proportional to pixel size squared, and inverse focal ratio squared.

 

All the best,

 

Kevin



#18 OleCuss

OleCuss

    Mercury-Atlas

  • *****
  • Posts: 2,655
  • Joined: 22 Nov 2010

Posted 07 November 2019 - 05:58 AM

Just to make sure things are not confused.

 

If one takes as an example the IMX071 sensor (it could be pretty much any sensor but being more concrete means we're not arguing about QE, pixel size, etc.)?  If you have a group of 4 pixels and expose for a period of time such that the group of 4 pixels detects 1,000 photons (an average of 250 photons per pixel). . .  If you bin those 4 pixels you will now have your software (or firmware) acting as though you have one pixel which detected 1,000 photons.  This means the number of photons and therefore the amount of signal was not increased.

 

Now you are probably going to want to process the group of 4 pixels with an average of 250 photons each differently than you will the one pseudo-pixel with 1,000 photons in it, but either way you still have a total of 1,000 photons.

 

And yeah, sometimes I do binning because I want the different processing or because I don't plan to be "zooming in" on an image and binning will thus better fit the image to the screen and/or because I like/want the different processing.  So I happily agree that there are reasons beyond noise reduction to do the binning.

 

But the only SNR benefit I can find to binning is noise reduction, not a signal increase.  In a low-noise situation the noise reduction may not be very valuable.

 

And then one caveat.  The ASI071 will actually do a sort of hardware binning.  It's a SoC function and is apparently fairly noisy so it actually increases the noise/artifact rather than reducing it.  So if the binning is done in a technically inferior manner it can actually increase the noise rather than decreasing it.  Beware hardware binning in CMOS cameras!



#19 Coconuts

Coconuts

    Viking 1

  • *****
  • Posts: 539
  • Joined: 23 Sep 2012

Posted 07 November 2019 - 07:54 AM

I disagree.  Larger pixels, whether they be physical or achieved through binning, decrease resolution but increase the etendue, and hence the imaging speed.  Etendue is a very basic optical fact of life.  I am forever asking (tongue in cheek) our optical designer to cut corners and violate conservation of etendue.  He refuses, not on principle, but on physics.  I'm afraid he's got me there.  In the OP of this CN thread is a link to an online etendue calculator (I contributed the variant on the right):

 

https://www.cloudyni...or/?hl= etendue

 

All the best,

 

Kevin



#20 OleCuss

OleCuss

    Mercury-Atlas

  • *****
  • Posts: 2,655
  • Joined: 22 Nov 2010

Posted 07 November 2019 - 10:17 AM

Because of the physics/engineering/design, one pixel large enough to fully occupy the space occupied by 4 (or 9 or whatever) smaller pixels should be expected to collect more photons.  Our microlenses just aren't that good.

 

But binning is a very different matter.  Invoking etendue does not create more photons.

 

Binning just doesn't create signal.  If binning creates photons then one should categorize those photons as "noise".



#21 Coconuts

Coconuts

    Viking 1

  • *****
  • Posts: 539
  • Joined: 23 Sep 2012

Posted 07 November 2019 - 04:37 PM

I again respectfully disagree.  The rate at which photons from an extended object (nebulae) accumulate in a given pixel is proportional to the square of the pixel's dimensions.  This is true irrespective of whether the larger pixel is created by lithography or binning.  A larger pixel looks at a larger solid angle of the nebula, and thus accumulates photons at a higher rate.  Spatial resolution is lost, but as long as this doesn't lead to undersampling, all is well.

 

Your comments "microlenses just aren't that good", and "large enough to fully occupy the space" isn't at all central to the above argument, but with the advance of BSI CMOS, microlenses are no longer necessary.

 

All the best,

 

Kevin


  • Jon Rista, donstim and bortle2 like this

#22 bortle2

bortle2

    Mariner 2

  • -----
  • Posts: 204
  • Joined: 18 Sep 2019

Posted 08 November 2019 - 02:13 AM

Because of the physics/engineering/design, one pixel large enough to fully occupy the space occupied by 4 (or 9 or whatever) smaller pixels should be expected to collect more photons.  Our microlenses just aren't that good.

 

But binning is a very different matter.  Invoking etendue does not create more photons.

 

Binning just doesn't create signal.  If binning creates photons then one should categorize those photons as "noise".

I know, it's counter-intuitive, and yet I'm sure astrophotography community will eventually come to the same conclusion photographic community did a while ago: with BSI CMOS sensors, pixel size does not matter; only sensor area does.

 

The change I'm referring to happened around 2015: that year, Sony released two cameras, A7R Mark II, and A7S Mark II, both FF. The former had 42Mp BSI sensor, while the latter had 12Mp conventional one, and was targeting video and low-light applications.

 

Sony's previous high-res model (A7R) had 36Mp conventional CMOS sensor, and initial knee-jerk reaction from the community was "42Mps is too much! Sony screwed it!! Who needs more megapixels if the image is ruined!!!111". Then 12Mp A7Sm2 was released... And people noticed that high ISO performance of the A7R2 was actually better than that of A7Sm2! I mean, really high ISO, such as 25,600. BSI made the difference.

 

Few remaining staunch proponents of "bigger pixels == better quality" fought their loosing battle for a while, but then even they gave up, because everyone could go to "Studio scene image comparison tool" on DPReview, pick A7Rm2 and A7Sm2, select ISO 25,600, and see by themselves that the former camera beats the latter for IQ, fair and square.

 

I'm sure the ramifications of the arrival of BSI sensors (for the "common wisdom" we're discussing here) will sink in in astrophotography community as well, we just need to give it some time...



#23 va1erian

va1erian

    Explorer 1

  • -----
  • Posts: 90
  • Joined: 27 Oct 2014
  • Loc: South Australia

Posted 26 February 2020 - 07:20 AM

Hello guys, when doing bin2 on a color ZWO camera, how the read noise, full well capacity, and dynamic range get affected? How the color arrangement is preserved?

I did the live stacking of M42 in bin2 mode, and the nebulosity came out very weak. In the bin1 mode, I could see the M42 extension. That is, bin2 made it worse not better. Maybe it has to do with debayering ? The camera is ASI 120MC-S.  



#24 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10,586
  • Joined: 03 Jul 2004

Posted 26 February 2020 - 10:54 AM

snip snip...

 

Few remaining staunch proponents of "bigger pixels == better quality" fought their loosing battle for a while, but then even they gave up, because everyone could go to "Studio scene image comparison tool" on DPReview, pick A7Rm2 and A7Sm2, select ISO 25,600, and see by themselves that the former camera beats the latter for IQ, fair and square.

 

snip snip ...

Let's see SONY engineers and marketing did their work or not...

(S is for sensitivity and R is for resolution)

 

ISO 3200

A7R2-A7S2_iso3200_.jpg



#25 ccs_hello

ccs_hello

    Voyager 1

  • *****
  • Posts: 10,586
  • Joined: 03 Jul 2004

Posted 26 February 2020 - 10:55 AM

ISO25600

A7R2-A7S2_iso25600_.jpg




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