There is often a focus on getting a suitable pixel size depending on the experiment, application or microscopy technique. An easy way to change sensor pixel size is to combine pixels into larger ‘superpixels’, also known as binning. This reduces resolution but increases sensitivity. Squares of pixels can be ‘binned’ into larger pixels, usually at 2×2 (combining a square of 2×2 (4) pixels into a single pixel) or 4×4 (combining a square of 4×4 (16) pixels into a single pixel), but can be done at much higher levels (even 32×32) depending on the limits imposed by the manufacturer. The effect of no binning, a 2×2 bin and a 4×4 bin can be observed in Fig.1.
It should be noted that binning differs depending on camera technology, namely charge-coupled device (CCD), electron-multiplied CCD (EMCCD) or complementary metal-oxide-superconductor (CMOS) sensors.
When a CCD/EMCCD sensor is binned it occurs on the sensor before readout, meaning that it occurs before read noise is introduced by converting photoelectrons (voltage) into grey levels (digital signal). The two primary benefits of binning with CCD/EMCCD is the improved signal-to-noise ratio (SNR) and the increased frame rate, but these both come at the cost of spatial resolution.
The SNR is improved as the read noise is only added to each superpixel. In a 2×2 bin, the signal is improved by four times (as each superpixel is four pixels), meaning the SNR is boosted by a factor of 4:1 (signal:noise). The frame rate is improved simply because there are fewer pixels to digitize.
As the CMOS sensor format is different to CCD/EMCCD sensors, binning also works differently. In CMOS the binning occurs off the sensor, after readout. This means that read noise has already been introduced to each pixel. Combining a 2×2 section of pixels together results in double the read noise for that resulting super‑pixel. Overall, it is four times the signal but two times the read noise, meaning only a 2:1 boost in SNR, compared to 4:1 in CCD/EMCCD.
Consider that there are only three major noise sources: photon shot noise, read noise, and dark current.
- Shot noise: If we detect 50 e- of signal, the shot noise is √50 or ~7 e-.
- Read noise: A fixed value for the sensor, for this case it is 1.7 e-
- Dark current: Image is taken with a short exposure, so dark current is effectively 0 e-.
This does not mean the total noise for this image is 8.7 e- (the total of the three noises). When adding noise together you take each value, square it, add the squares and take the square root of the result. This is called ‘adding in quadrature’. Our equation looks more like this:
The noise in this situation is dominated by shot noise, the effect of read noise and dark current is negligible. This results in an SNR of 50:7.2 or 7:1. But while the noise is added in quadrature, the signal is added normally. If we perform a 2×2 bin on four pixels that each have the same signal of 50e- and same noise values, we have a signal of 200 e- (50+50+50+50), but noise of:
After a 2×2 bin, this gives us a new SNR of 200:14.4 or 13:1. Compared to the 7:1 before binning, the 13:1 after a 2×2 bin only results in double the signal, whereas this would be quadruple in CCD/EMCCD. However, the SNR has still increased, meaning that while binning is not as effective in sCMOS, it can still boost the signal, albeit with no speed change and a decrease in resolution.
Binning allows researchers to increase pixel size in order to gather more signal at the cost of resolution, increasing the flexibility of a camera sensor. Due to the manner in which different camera sensors work, it is important to note the differences in binning between CMOS and CCD/EMCCD.