In digital photography noise is typically associated with the mottled image we get when taking shots in the low light with a high ISO. Noise was an issue with film cameras as it is these days with digital cameras—its presence is not only due to the fact that with modern cameras we are trying to convert an analogue signal to a digital one.
There are, in fact, several different types of noise—but they can broadly be placed into two categories: photon (owing to natural random fluctuations) and electronic (sensor and circuitry related). Photon noise, read noise (electronic), and thermal noise (electronic) are the main types of noise in digital imaging.
Photon noise is responsible for the majority of noise we see in photographs. It results from the fact that light arrives in photons—discrete bundles of energy—which are subject to random fluctuations. We speak of an average flux—the average number that arrive in a second over a given area (the area of a pixel, say). The more intense the light the more that arrive in a given time period. The issue is that this is an average—and varies both spatially and temporally. In other words, even with a constant light source the exact number of photons arriving in the pixel may vary from second to second and from pixel to pixel. Even if you had the world’s most perfect sensor photon noise would still be present since it is a result of the physical properties of light.
Imagine, for instance, if we placed three cups outside in torrential rain for one minute. We would expect them to fill up at more or less the same rate but we couldn’t guarantee that each cup would contain the exact same quantity of what once the minute was up.
Now the variation observed follows a normal distribution (bell curve) and so the standard deviation—the average fluctuation from the mean average (noise)—is equal to the square root of the average photon count (the “signal”). For instance, if on average 10,000 photons fall on the pixel during the time of exposure then the standard deviation would be 10,000½ = 100 and the typical reading would range from 9,900 to 10,100. If the light intensity is reduced to only 100 photons then the range is now 90 to 110.
The signal-to-noise ratio is thus given as follows:Consequently, the higher the signal (the greater the light intensity) the higher the signal-to-noise ratio. This is why for a given exposure, a longer shutter speed or greater aperture at a low ISO is will give less noise than a fast shutter speed or smaller aperture with a high ISO—more photons will be collected in each pixel!
Steps to dealing with noise have come along in leaps and bounds in recent years—so much so that we speak of higher end DSLRs being able to “see in the dark”. The main reason that DSLRs (in particular full frame DSLRs) are so good at dealing with noise relative to your standard point-and-shoots is because the individual pixel size is much larger. Consequently they are able to capture many more photons, like a bucket versus a cup left out in the rain. Similarly, it is the reason that the mega-pixel battle in consumer camera market does not lead to better quality images—just a bigger number to grab people’s attention in the shop.
Electronic noise refers to noise which is produced as a result of imperfections and random variations in the cells and electronic circuitry which ultimately turn photons into a digital signal. The diagram below shows the flow between photon and digital signal.
1. Photons hit the photodiode
Ideally, the photodiode would emit one electron for each photon that arrives, but this is not the case in real sensors—some photons do not register. The success rate of converting photons to electrons in a photodiode is called the quantum efficiency and for digital cameras this figure is usually within a range of 30-50% (see here for a list of quantum efficiency ratings).
2. Electrons stored in a well as an electric charge
The electrons that are emitted by the photodiode are collected during the exposure time and stored in a well as an electric charge. The well will have a maximum storage capacity beyond which the pixel will register as pure white and, in some cases, may result in blooming, where charge overflows into surrounding pixels.
3. Amplification and conversion to Analogue-to-digital units (ADU)
Once the shutter is closed, the charge in the well is released and amplified depending on the ISO setting. This amplified charge is then converted into a digital unit. The relationship between the number of electrons and the number of analogue-to-digital units is called the gain (expressed in electrons/ADU or e-/ADU). Gain halves as we double the ISO number. So, for instance, if the gain at ISO 100 is 7.6 e-/ADU then at ISO 200 it will be 3.8 e-/ADU and so on. The term gain is, in a sense, a bit of a misnomer given how it is expressed (inversely proportional to the ISO); however, it makes more sense when you think of it in terms of amplification of the initial signal which is proportional to ISO.
Remember, the ultimate aim of the sensor is to convert the photon reading into an ADU. For a 16-bit camera this means assigning a value between 0 and 65,535 to each pixel. This is the RAW value from which the camera (or computer software) can run its algorithms to assign a colour.
In a perfect sensor, if the same number of photons arrived at two different pixels the final ADU output would be identical. Unfortunately, this is not the case. Minor variations in quantum efficiency and voltage fluctuations at different points in the circuitry mean that there are deviations from photon values to final ADU output. This is referred to as read noise and is typically the main source of electronic noise.
Another culprit, thermal noise, exists because of the very small electrical current that flows through the photodiodes even when no photons have reached the sensor, called the dark current. This can release electrons to the well and is an unwanted signal which increases with exposure time (hence dark noise is primarily a concern for astrophotographers). In its most extreme form it can lead to hot pixels—pixels whose wells have become fully saturated with electrons due to the presence of dark current.
Technically speaking, the electrons produced because of dark current are not noise but unwanted signal. And because this unwanted signal is a repeating pattern it can be addressed in-camera by taking a dark frame (an exposure without opening the shutter), assessing the pattern of the unwanted signal and deleting it from each picture. However, the unwanted signal itself is subject to random variations so cannot be entirely eliminated—and what remains is, technically speaking, noise.