ISO (light sensitivity)

ISO stands for International Organization for Standardization. The standard that regulates the light sensitivity of digital cameras is ISO 12232:2006. It is common to use ISO for short when discusing the light sensitivity of cameras.

How to use

ISO is usually specified on a logarithmic scale like 100, 200, 400, 800. Each step doubles the light sensitivity of the camera so ISO 800 is eight times as sensitive to light as ISO 100 making it possible to use 1/8 of the exposure time. It is quite common for cameras to have itermediate steps for finer adjustment.

Don't be afraid to increase the ISO. A good picture is still good if it is grainy but a picture with unintentional motion blur is always a ruined picture. Remember that correct exposure becomes much more important at higher ISO levels. If you underexpose at ISO 100 there is still alot of information ready to be lifted out of the dark areas but at the highest ISO setting there is nothing more to gain by increasing the brightness, what you see is what you get.

How it works

The common way to adjust sensitivity in a digital camera would be like this:

Pixel --> variable gain analog amplifier --> analog to digital converter --> digital multiplier

Each colour channel have a separate gain setting which would be part of the white balance system.

The digital multiplier is often referred to as uncalibrated or HI setting. Usually it multiplies the pixel value by a power of two, the same effect can be had by adjusting the brightess with image processing software. There may or may not be an advantage of letting the camera do it because it may affect how the noise reduction algorithms in the camera works. It needs to be tested on every camera model to find out what works best.

The number of photons that hit a pixel is a digital value so in some sense the analog amplifier is not analog at all. When the number of photons is large the quantization of values is completely hidden in noise and the signal seems analog, but as the number of photons decrease there will discrete values. For example if each pixel registers 0 or 1 photon the image is digital from the very moment the shutter closes.

The noise is determined mainly by how many photons each pixel registers. Since the photons spray randomly from an object it will hit the pixels in the camera in a random order. If the scene is completely evenly and brightly lit, each pixel will still contain significant random noise just from random chance. If each pixel registers 65536 photons on average the average noise would be sqrt(65536) = 256 but there is nothing to stop some pixels to be completely black by random chance, it is just extremely unlikely for it to happen.