With the introduction of the Atik Horizon, our first CMOS camera, it’s a good time to look at the differences between CMOS and CCD technologies. One area where they differ is how they handle binning. This article will touch on the main differences between CMOS and CCD binning and explain the implications of these.
So what is Binning?
To put it very basically, in both CCD and CMOS, “binning” is the technique of combining pixels in an image into a “super pixel”. It typically takes place in 2×2 or 3×3 blocks etc. In doing so, the signal in the individual pixels is combined into the super pixel. This helps improve the signal to noise ratio (SNR). However, it also lowers the resolution of your image as it contains fewer, but larger pixels. Where CMOS differs is that this process takes place in the software after the digitalisation of the pixels has been completed. This is very different to CCD where binning takes place in hardware, or “on-chip”.
To help visualise this process, below we have 4 pixels (blue) and each pixel contains one electron (red).
When binning pixels, we add their values together to create a single pixel which holds all 4 of the electrons.
So that’s 2×2 binning explained in a rather basic way. To help clarify, below is an example of how 2×2 binning would take place with 16 pixels. They become 4 super pixels while still maintaining the equivalent amount of electrons as their un-binned counterparts.
To summarise what binning is:
- Combining adjacent pixels, most commonly in a square grid (2×2, 3×3, 4×4)
- Pixels are combined into a “super pixel”
- Binning improves SNR, but at the cost of resolution
- CMOS binning takes place in software, after digitalisation
- CCD binning takes place “on-chip”, before digitalisation
How do CMOS and CCD binning differ?
One of the main reasons for binning is to detect faint signals. The detection limit is often defined as the point where the SNR is one, meaning the signal is equal to the noise. In the case of faint signals, the noise is just the read noise of the camera. So if we take a read noise of 3 electrons (e-) and a signal of 3e-, we have a SNR of 1 and the object will be just about detectable.
CCD technology allows pixels to be combined in the analogue domain, before they are read out from the sensor. So taking the above example with a 2×2 super pixel, it will have the signal from all 4 individual pixels; 4x3e-=12e-. As we only read the super pixel once we still only have 3e- of read noise. So the SNR ratio is now 12/3=4. In other words, 4 times better.
CMOS technology is different in that the pixels cannot be combined into a super pixel before they are read out. What we can do is combine the pixels mathematically after readout on the computer. Each super pixel again receives the signal values from its previous individual pixels, but because this is happening after the pixels have been read out, there is also an increase in read noise. However, the read noise increases at a lesser rate than the signal, giving us an improved SNR. Let’s take a look at some of the maths behind it.
Staying with the above example, the signal is again 4x3e-=12e-. However the individual pixels have already been digitised and each have 3e- read noise. As previously mentioned, noise in an image doesn’t accumulate as a normal addition but follows a statistical distribution as the square root of the number of pixels added. So in this case the noise of our CMOS super pixel, this is 3e-x4/sqr4=6e-. So now our SNR is 12/6=2, twice as good as without binning but not as good as with CCD technology. As the amount of binning increases, 3×3, 4×4 etc, the difference between signal to noise performance in CCD and CMOS also increases.
CCD binning is a very powerful technique allowing increased sensitivity at the expense of resolution. This has a number of applications:
- In short exposures to find if an object is centred in the picture
- Capturing RGB data in LRGB imaging
- Matching a camera to a long focal length telescope
- Imaging faint objects
Binning is less effective in CMOS cameras. The read noise of a CMOS sensor will tend to be lower than CCD which helps with lower levels of binning, but the fact it includes the read noise of every single pixel really adds up as the amount of pixels you bin together increases. Other processing techniques such as noise reduction, resize and resample may be better at extracting the best signal from CMOS data.