Binning Explanation

[Fay]

Can someone do a quick explanation of binning & when it would be used & why. I seem to remember that  Chris uses this a lot for RGB.
If so, when combining, would you not combine a 1x1 L with 2x2 RGB? Would you have to reset the RGB to 1x1?

Thanks,

Fay

[RobertM]

Hi Fay,

The primary benefits of binning are to improve the signal to noise ratio for chips where readout noise (primarily for Kodak) is significant or to match resolution to image scale.

The reason that you would use binning for RGB is that the eye is less sensitive to detail in information of individual colours.  So in theory you can use binning to increase the 'effective' sensitivity of your sensor in those colours i.e. for a particular exposure length you will be collecting four times more photons per cell binned 2x2 pixels than 1x1 because that cell spans a 2x2 matrix of ccd pixels.  Because of this you can either use subs that are a quarter the length or go deeper (as you are collecting 4x the photons per cell).  It is important to note that you will still be collecting the same number of photons unbinned as binned as the collection area on the chip is the same.

It is also important capture a high res image at all wavelengths (luminance) to go with binned colour as this provides the detail for the image.  The low res binned colour imformation plus the high res unbinned luminance fools the eye into thinking that the whole image is in high res colour.

To merge the images one set would have to be rescaled to the other.  As an example if you had a sensor or 640 x 480 unbinned which when binned 2x2 would be 320 x 240 then to merge the images that would have to be rescaled back up to the full res of the sensor.  You could do it the other way around but then you would loose resolution.

For my own part I am not happy with this approach, my reasons are as follows:

1) By binning you loose resolution which can never be recovered.
2) The binning process can be done in software after the image has been downloaded in Maxim or some other program - after all the binning process is just a summing of adjacent pixels in firmware on camera/chip anyway.
3) Starlight Xpress cameras have very low read noise so I don't think it's necessary.

On the plus side, you do get to see the image quicker as a 2x2 is four times more sensitive so the subs can be 1/4 the length, the images take up much less HD space and are quicker to process due to the smaller file sizes.

Fo me important reasons to use binning would be:

    Focusing is quicker (faster downloads) and shorter exposures.
    For framing the image on the chip.

There are plenty of explanations on the web, and no doubt others have their own ideas on this topic.

I'll see if I can remember to bring along a CD to the next meeting with lots of imaging tutorials on.

Hope this helps.
Robert

[Fay]

Robert, that was very helpful. Thanks for putting so much effort into your answer.

[Rocket Pooch]

Fay,

It like a bucket (pixel=bucket), if you have a small bucket (1x1 binning) you capture less rain (photons = rain), a bigger bucket (2x2) more rain.

If you do your binning after imaging, you are taking the 4 small buckets and pooring them into one big one.  Which is higher but not wider than the origional bucket, therefore the bigger buckets take less area than the smaller ones and have more rain in them but over a smaller area.  Which is why the images are smaller.

Thats about it, see complex computer drawn image below, it took me bloody ages and I had to ask Frodo how to do it.

r = rain

x = bucket edges

n = bucket noise



un-binned

x     x     x  r  x     x      x     x
x     x  r  x  r  x  r  x      x     x
x  r  x  r  x  r  x  r  x   r  x     x
nnnnnnnnnnnnnnnnnnnnnnnnn



binned

x     x     x     x
x     x  r  x     x
x     x  r  x     x
x  r  x  r  x     x
x  r  x  r  x     x
x  r  x  r  x  r  x
nnnnnnnnnnnnn
nnnnnnnnnnnnn

Chris

P.S. Can someone add why we should binn when imaging and not after, i.e. the relationship between noise in image a above and image b?

[MarkS]

P.S. Can someone add why we should binn when imaging and not after, i.e. the relationship between noise in image a above and image b?

Each photon leaves one electron in the bucket.  An amplifier has to count them - but it can't count them exactly because of amplifier noise.  This extra noise, which is introduced when reading the pixel value, is called read noise.

If you read 4 small buckets then each of the 4 numbers will have a noise component introduced.  But if you do hardware binning then only one noise component is added to the total because all 4 bins are counted in one go.

The net effect is that hardware binning has half the noise of binning in later in software.

There are two other types of noise:
1) Thermal noise
2) Poisson distribution noise (photon noise)
But these two types of noise are not affected by whether you do the binning in hardware or software.

[RobertM]

Reduction in read noise is the only advantage I can see.

I see one disadvantage as being that the internal registers in the ccd may not be able to cope with summing the charge of adjacent pixels when nearly full.

[Fay]

Certainly a lot of info to digest, I will have to read it all a couple of times to take it all in.

Chris you have answered a lot of posts today!!!!!! I can thank Frodo for the wavelets explanation!!!!!

[Rocket Pooch]

I had a chat with Frodo he said if there's anything needed in relation to hair, licking things and also chasing balls then he can answer. 

He is also currently working on the relationship between gravity and spherical objects when a force is applied to the sperical object which will make it travel along a determined line.

Mark & Rob good answers, lets have a pop quizz wiki.

[RobertM]

Was the bucket and rain analogy something that Frodo thought up the whilst the spherical object (apple maybe) was travelling along that determined line from the branch to his head or after it hit ?  The answer to that question could compromise any application the Royal Society !

[Rocket Pooch]

No its ok the sperical object was a sponge ball.