Software Binning

Comments

• BlockHead

  January 2022

  Hi Bruce,

  
  

  It gets a little complicated... because you can't just bin (or resample) calibration data willy-nilly. 

  I have a couple of comments.... 

  
  

  1. Hardware binning is the only way to *benefit* from the reduction of read-noise. This is a real effect that improves the S/N of images for some sensors. There is *no* particular benefit to binning after-the-fact unless you are matching 2x2 with 1x1 acquired data. 

  
  

  2. Yes, IntegerResample is the proper tool to do this operation. You do not want the additional interpolation that potentially comes with the regular resample. 

  
  

  3. The choice of method depends on whether it is important to conserve flux. And this answer depends on what you are resampling. 

  
  

  4. When you use it depends on the usage... this is a bit complicated! I do not think you want to do this right after data collection..and I positive this will make a mess of the calibration of the data. So I do not understand what the point of this is. In my mind, you have to calibrate the data first...which means you need to acquire and calibrate the data from the chip at 1x1 and then resample... or you put the chip acquisition at 2x2 and calibrate. 

  
  

  I think you need to spell out why you want 2x2. 

  
  

  -the Blockhead
• Bruce Donzanti

  January 2022

  Hi Adam

  
  

  My main reason for asking if and how I should do this is that my setup under my usual sky conditions would indicate I am oversampling.  So, I was assuming (maybe incorrectly) that bin 2 would help produce a "better" image.  If I understand things correctly, bin2 would boost the S/N at the expense of some loss in resolution.  So, I was trying to decide if that trade-off leads to a better image or not.

  
  

  I've attached an example at bin 1 of a region of the Rosette taken under my usual sky conditions (Bortle 7, OK seeing).  Taken with a C11" EdgeHD @f/10 with Chroma ha (5nm) and OIII (3nm) filters at 300s exposures using a ZWO ASI6200MM (3.76 pixel size), 0.27 arsec/pixel.  In essence, without mincing words, is this a poor image and would binning 2x2 potentially enhance it?

  
  

   

  
  

    

  

  

  HOO Final.jpg

  9092 x 6014 - 19M
• Bruce Donzanti

  January 2022

  And the same image in Ha only...........

  

  

  rosette_Ha v2.jpg

  9576 x 6388 - 15M
• Wade Hilmo

  January 2022

  Hi Adam,

  
  

  Regarding your comment above about hardware binning and read noise, that applies only to CCD cameras.

  
  

  CCD cameras have a single analog-to-digital converter and they run all of the pixels through it during read out.  When you bin a CCD camera in hardware, the pixel combination happens before the A/D converter, so the read noise hit happens after the pixels are combined into "super pixels".

  
  

  CMOS cameras are completely different in the way that they handle reading out the data.  With a CMOS camera, each pixel has its own integrated electronics, including A/D converter.  As such, it's impossible to combine pixels before A/D conversion.  With any CMOS camera "hardware" binning is actually done in software - I believe by the driver.  But since the binning can never happen before A/D conversion, there is no read noise benefit to binning with a CMOS camera.

  
  

  My guidance for CMOS data is to always capture the data with the camera binned 1x1, even if the data is significantly over sampled.  Some processes (like deconvolution) even benefit from working with over sampled data.  I do a resample of CMOS data near the end of processing to bring the resolution in line with something appropriate to the data.

  
  

  The exception to this, is that some of the newer CMOS cameras have incredibly large numbers of pixels that can produce quite large files.  If your pixels are small enough that you have lots of "empty resolution", it may be worth having he camera driver do the binning so that the file sizes are smaller off the camera.  Even this varies from camera to camera, though, since some CMOS cameras (the ASI1600 comes to mind) reduce the number of bits per pixel when you bin them.  For a camera like the ASI1600, you would never want the camera to bin in the driver. 

  
  
• Bruce Donzanti

  January 2022

  "My guidance for CMOS data is to always capture the data with the camera binned 1x1, even if the data is significantly over sampled.  Some processes (like deconvolution) even benefit from working with over sampled data.  I do a resample of CMOS data near the end of processing to bring the resolution in line with something appropriate to the data." ........This gets at the heart of my question.  I have done deconvolution, but frankly, I don't notice much of a change, with a slight reduction in star size and even the nebula giving a very small apparent boost in clarity.  Perhaps the slight change is all that is supposed to happen or maybe my data are not as oversampled as I believe.  Thus, that is what led me down this road of 2x2 software binning.   
• BlockHead

  January 2022

  Yes, Wade is correct concerning the CMOS cameras. I forgot to add this point. This is what I mentioned the only benefit in terms of actual signal improvement is through on-chip binning... AND you cannot take advantage of this with CMOS sensors. Sorry about that.

  
  

  Jon Talbot's presentation explains one facet of this in his AstroImaging Channel appearance. 

  
  

  
  

  So to Bruce's point... you may be oversampled- but there isn't anything you can do with the sensor to improve the sampling. It is the sensor that does this. Doing something after the data is collected (counted) by the hardware doesn't materially affect things. You can certainly add a focal reducer or purchase a shorter focal length telescope. 

  
  

  Concerning the deconvolution question- adjusting binning (even if it were possible) to address deconvolution woes isn't the best route. (Something like cutting off your nose when you have a cold.)

  So if this is about deconvolution, a different conversation about how to make it work (or why it isn't working out) is really the topic at hand. 

  
  

  Concerning "not as oversampled"... to put a number on it, a FWHM between 4-6 pixels is usually about right. The data being deconvolved also needs to be near gaussian and bright (say 0.05 to .1 and greater).

  
  

  -the Blockhead 

  
  

  
  
• Bruce Donzanti

  January 2022

  Thanks for the video link.  Based on this, the other information provided here, and other readings I have done in the past, including asking others who are using this camera, there really seems no benefit to 2x2 binning that I can tell other than smaller file sizes.  Using a reducer is a good idea but unfortunately, it is poorly designed for this scope. 

  
  

  I have watched your videos on deconvolution and tried it.  I guess I was expecting a more dramatic effect, but you do explain the effect is NOT dramatic.  

  
  

  Thanks for the numbers on "not as oversampled" and deconvoluted data as I obviously was guessing at all of this.