ISO in Digital Photography
This short tutorial explains how digital ISO is determined, and how setting higher and lower ISO speeds work in a digital camera.
Note that the actual documents describing the criteria used for determining ISO on film and digital sensors are copyrighted. If you want to read them yourself, you must purchase them through the ISO website. In this article, I shall just summarise some of the most important principles in determining the ISO of a digital sensor. (Note: Kodak application note MTD/PS-0234 summarises some of the material in the ISO standards, and can be downloaded for free.)
If you've used film, you've probably noticed the number stamped on the outside of every box to indicate how sensitive the film is to light. This number is known as the film's “ISO”. The higher the ISO-number, the more sensitive the film is to light.
In digital photography, the ISO plays a similar role. That is: The ISO-number is a measure of how sensitive the camera's digital image sensor is to light. The same principles apply as in film photography: The higher the number, the more sensitive the sensor is to light.
The ISO of a particular film and the ISO of a particular digital sensor are set by the manufacturers of the film or the digital sensor. The International Organization for Standardization, which also is known as “ISO” has created a set of criteria for the manufacturers to use when they determine the ISO of a particular film or digital sensor. These standards aim to provide a consistent framework for how these numbers are determined. For digital cameras, the standard defines ISO speed in such a way that a digital sensor and a film that both are rated with the same ISO speed will behave in a similar way when exposed to the same amount of light. This means that light meters and exposure techniques that works well with film, also will work when the photographer uses a digital camera.
A Brief Historical Interlude
The sensitivity scale defined by ISO actually defines two parallel scales, one linear (arithmetic) scale and one logarithmic scale. This is because the ISO sensitivity scale was created in 1987 by merging two older scales known as “ASA” and “DIN”. The ISO linear scale corresponds to the older “ASA” scale, and the ISO logarithmic scale corresponds to the older “DIN” scale.
When both values are written, they should be separated by a slash (/) and the logarithmic value is marked with a degree (°) symbol. Example: “ISO 200/24°”. If only one value is written, it is always the linear value. Example: “ISO 200”.
In the ISO logarithmic scale, adding 3 to the numeric value indicates a doubling in speed, while the ISO linear scale, doubling the number indicates a doubling in speed.
In practice, the ISO logarithmic scale is no longer used, and you will instead see the film or sensor speed measured in the linear values. From this point on in this article, I shall ignore the ISO logarithmic scale and only refer to the linear scale.
As we have already seen, the linear ISO scale is arithmetic and double the linear ISO number means double sensitivity. In other words, a film or sensor with a speed of ISO 200 is twice as sensitive to light than a film or sensor with a speed of ISO 100.
ISO Speed and ISO Latitude
In the context of sensitivity to light, there is actually two ISOs: ISO speed and ISO latitude. The former indicates a specific sensitivity to light, and the unit of measurement is simply known as “ISO”, the latter designates the range of ISO speeds that will produce “acceptable” images with a particular film or digital sensor.
It follows from ISO latitude that the ISO speed can be changed within a certain range.
With film, ISO speed is changed by altering the chemistry of the developer, the agitation, the temperature and/or the time of development. If the ISO speed is increased, this is known as “pushing” the film, if the ISO speed is decreased, this is known as “halting” the film.
With digital, ISO speed is changed by analog or digital amplification. This amplification and its consequences will be discussed in more detail below.
Before we move on to digital, I shall briefly mention how the ISO-number for a particular film is determined.
The principles for determining the ISO for film is described in three separate documents ISO 6:1993, ISO 2240:2003 and ISO 5800:1987. They cover black & white negative film, colour reversal film and colour negative film, respectively.
The procedure described involves exposing the film in question to specific amounts of light, then giving each exposure a specified degree of development in a specified developer. The point in the sequence where a specified minimum density fogging occurs, determines the ISO speed the manufacturer shall put on the box of film. This is the native (or base) ISO speed of the film.
By halting or pushing the film in development, the ISO speed can be decreased or increased. Halting development reduces contrast while increasing the film's dynamic range. Pushing film has the opposite effect, and also makes the grain larger and more visible.
Films exposed to light and developed produce fogged images that can be read with a densitometer. This reading determines the ISO for a specific film. Unfortunately, this method can not be used to determine the ISO of a digital sensor. Digital ISO must therefore be determined by methods different from those used to determine film ISO. The methods for determine ISO are described in detail in the following document: Photography – Digital still cameras – Determination of exposure index, ISO speed ratings, standard output sensitivity, and recommended exposure index; ISO 12232:2006.
Like film, digital image sensors are said to have an innate “native” (or “base”) ISO speed. This is usually ISO 100 or ISO 200 for modern sensors. It is supposed to be the sensitivity where the sensor gives its “best” performance, as determined by dynamic range (large) and noise (minimal).
Canon has never published the native ISO speed of the sensors used in their DSLRs. Users, however, have determined that some of Canon's earliest DSLRs had the largest dynamic range and lowest noise levels at ISO 100, while later models perform best in these respects at ISO 200.
Nikon used to publish the native ISO speed of the image sensor used in their DSLRs as part of the specifications. ISO speeds below this were indicated as L0.3, L0.7 and L1.0 respectively. For most models Nikon indicated that the native speed was ISO 200. With some recent DSLR models (D3100 and D7000), Nikon no longer offer “L” settings and the lowest ISO speed is simply ISO 100. Looking at the dynamic range of these cameras, it is lower on ISO 100 than ISO 200. To me, it looks as if Nikon for some reason has copied Canon and no longer report the true native ISO speed as the lowest ISO setting in the camera's specifications.
In digital cameras, the conversion of light into bits works like this: The electronic sensor collects photoelectrons (produced by photons colliding with the image sensor and knocking electrons loose from the valence band) into photon wells. This process accumulates charge in the photon wells. The amount of charge that has accumulated in a photon well indicate the amount of light that has illuminated the photo well since the well was reset.
The accumulated charge from each photon well can be converted into a voltage. This voltage is fed into an Analog-to-Digital Converter (A/D Converter) to produce the RAW digital data bits that (eventually) will result in the digital pixels that make up a digital image.
The diagram above shows how the digital output from a 12 bit A/D Converter may look like as a function of the number of photons that hits one photon well during the time of exposure. The left of diagram shows the output when there is no light (zero photons hitting the sensor), the right shows very bright light (one million photons hitting the sensor). The blue area shows the signal and the red area shows the noise. Notice that both axes are logarithmic.
Photon wells behave in a linear fashion, but have limited capacity. When a photo well has accumulated the maximum charge it can hold, indicated with a green line in the diagram, supplying more light will only result in non-linearity (burnout), and maybe also spillage into adjacent pixels (blooming). The native quantum efficiency (the percentage of time an incident photon actually results in a photo-electron) and the charge well capacity determine the so-called native ISO of the sensor.
In addition to the signal, some noise is also produced by the process. This is a combination of so-called “dark-current” noise (current always present, even when no light is hitting the sensor), readout noise, and thermal noise. This is indicated by the red area. In the darkest shadows (to the left in the diagram), some noise is present, and masks the signal. As the signal becomes stronger, it overpowers the noise. The output values where the noise masks the signal, are said to be below the noise floor.
Note that the linear behaviour of a photon well is cut off by abrupt non-linearity at both ends. It is cut off by the noise floor to the left, and burnout to the right. This is very different from film. Film's response to incoming light has the shape of an S-curve. This means that film is much more forgiving for both underexposure and overexposure than a digital sensor. RAW-converters usually defaults to an S-shaped tone-mapping curve to emulate the response of film, but tone-mapping cannot bring back the signal that is lost in the non-linear areas of the response curve (i.e. when the signal is below the noise floor or beyond full well capacity).
Dynamic and Tonal Range
The dynamic range of a digital sensor is determined by the difference between the point where the noise masks the signal, and the point when the sensor is no longer linear because full well capacity is reached. The tonal range of a digital sensor is determined by the number of usable bits representing luminosity values.
Dynamic and tonal ranges are related, and are sometimes referred to interchangeably. However, as is illustrated in the figure below, they are independent. We can have image data that have a high or low tonal range, and also image data with a high or low dynamic range, and any combination of the two.
The dynamic range is a measure of how much the darkest bits in a recorded scene differs from the lightest. It is usually expressed in EV, where an increase in luminance equal to 1 EV representing a doubling of the light.
The tonal range is a measure of the granularity we use when real world tones are mapped onto an recording medium. With a high tonal range, gradients are smooth. With a low tonal range, the gradients are abrupt, and we see an image defect usually called banding.
Adjusting Digital ISO
While film ISO can be changed (by pushing or halting development), every frame on a roll of film must be exposed to the same ISO to avoid under- or overexposed images.
As every owner of a digital camera knows: The ISO on a digital camera can be adjusted on a frame by frame basis. In fact, most cameras let the photographer set a ISO speed that is both lower and higher that the camera's native ISO speed. When you do, the camera also adjusts its internal processing to match the new ISO speed. With a digital camera, you can use the appropriate ISO speed for the scene on every frame.
When you set a higher ISO speed on a digital camera, the camera's light meter will use this ISO value as the basis for making its measurements. Also the digitized RAW data will be adjusted. (Usually by having amplifiers in the image sensor's circuitry increase the gain before sending the analogue voltage read from the photon well to the A/D converter to be digitised. E.g.: If at ISO 100 the signal is 100 mV, we can get ISO 200 by using an amplifier to boost it to 200 mV. For ISO 1600, we can five-double it to 1600 mV, and so on.)
The idea behind letting the photographer adjust ISO is mainly to ensure that the full input voltage range of the A/D Converter is utilised. This means that the full bit depth of the A/D Converter converter is used, so we get the maximum tonal range from the A/D Converter. However, the sensor's sensitivity doesn't actually increase; the camera is just amplifying the data it produces. When we amplify the signal, we also amplify any noise, so we will lose some image quality. This is similar to what happens when you “push” film speed, where grain also degrade the image.
While setting a higher ISO speed on the camera does not change the base ISO of the sensor, doing so it still impacts on the RAW data emerging from the A/D converter. The exact behaviour varies from model to model, but in most cameras, raising the ISO will result in less noise and a greater dynamic range than shooting at base ISO and increasing the exposure by means of software.
The pair of images below show two shots with a Nikon D700 using the same aperture an shutter settings (f/5.6, 1/40 second). The image on the left is taken with the camera set to ISO 3200 and unprocessed (i.e. automatic converted from RAW to JPEG to in the Adobe Camera Raw (ACR) CS4. The image on the right is taken with the camera set to ISO 200, which is the base ISO of the Nikon D700 (i.e. deliberately underexposed -4 EV, and the exposure is the lifted +4 EV using ACR CS4). Capture scaled to fit on the screen on top, 100 % crops below.
The pair of images demonstrate that on a Nikon D700, adjusting the ISO to make the exposure “right” results in a less noisy capture than staying at the base ISO.
The grain introduced when you push film to higher ISO speeds is somewhat more pleasing to the eye than the digital noise that appear when you boost the ISO speed of a digital sensor. But the latest generation of digital sensors have been very good in controlling noise. As a result, most modern DSLRs produce better image quality at high ISO speeds than film do.
Lowering the sensor's ISO below the native ISO is not really practical. Due to popular demand for low ISO settings, most cameras let the photographer do this, either by just providing lower settings, or by calling such a setting “L0.3”, etc. to indicate that it is off the scale. However, this is not recommended, as using it will result in some loss of dynamic range.
Here is the explanation: When you set a lower ISO speed than the sensor “native” ISO speed on a digital camera, the full well capacity is still determined by the number of incoming photons, but the signal is lowered to simulate a lower ISO value. This does not produce more well capacity, nor does it do anything to the noise floor, so the net result is that the sensor have a lowered dynamic range and the same noise level as it has at its “native” ISO speed. In other words, there is nothing to be gained from using an ISO speed below the sensors “native” ISO speed.
The ISO speeds you can set on a specific digital camera is selected by the manufacturer so that they will provide the correct exposure required to produce an image of specified brightness similar to what a film with a similar ISO speed would produce, while restricting image noise to a specified acceptable level. For the native digital ISO speed, that is usually the lowest possible noise level.
For ISO latitude, the upper ISO speed is determined by a higher but still acceptable specified noise level (“noise limited”), while the lower ISO speed is determined by how much highlight clipping and reduction in dynamic range the manufacturer thinks is acceptable (“saturation limited”).
Because exposure duration, temperature and humidity can affect digital image quality, the ISO standard cites specifics for each of these. In real life, we shoot at a wide range of shutter speeds, temperatures and humidities so, as is the case with film speeds, the controlled laboratory criteria don't necessarily apply to the wide range of real-world photographic situations. But like ISO film speeds, digital ISO speeds provide us with a standard starting point.
Using a Wider A/D Converter
For a long time, 12 bits has been the norm for A/D Converters in digital cameras. Instead of analogue amplification, it is possible can use an wider A/D Converter converter with higher than required bit depth (e.g. 16 bits) and just rescale the output data digitally afterwards. One bit is lost for each multiplication by two (1 EV). If we have a sensor with ISO 100 as native ISO, using a 16-bit A/D Converter would allow us to crank the speed up to ISO 1600 and still output the standard 12 bits per photon well, without the need for any analog amplification. Some cameras use this technique, alone, or in combination with analogue amplification, to provide ISO speeds above the native ISO.
Using a 16 bits A/D Converter does not remove the noise, because digital multiplication also magnifies noise, just like analogue amplification does. However, having an unamplified digital signal allows for alternative signal processing techniques to remove noise from the data stream, and some argue that this yields better noise removal at high ISO speeds.
Digital ISO and Noise
Just as some ISO 400 colour slide films produce better image quality than others, some digital cameras produce better image quality at a given ISO setting than others. In general, a large photon well produces a strong signal (resulting in a good signal/noise ratio).
In the illustration above, the collection of photon wells is illustrated by an analogy: Rain drops being collected by buckets. An empty bucket corresponds to black, a full bucket corresponds to white. When making an exposure, the shutter works as lid allowing the buckets to collect drops for some time. By measuring the level of water in the bucket that has accumulated in the interval the lid/shutter is open, we can determine the exact shade of grey between black and white to assign to the particular bucket/photon well.
As shown, both the small bucket and the large bucket will be filled with water at exactly the same rate. But the larger the surface of the bucket, the more rain drops will be collected in a given amount of time.
When collecting photons, a sensor cell with a small surface area corresponds to a small bucket, and a sensor cell with a large surface area corresponds to a large bucket. The level of the signal is proportional to how high the water level inside the bucket is, and will be the same for a small bucket and a large bucket. But the accuracy of the signal is measured by the signal/noise radio (the signal level divided by the noise level). A larger bucket/photon well will collect more drops/photons in a given amount of time. This means that a large photon well will produce a stronger signal than a small photon well, resulting in a better signal/noise ratio for the large photon well.
The physical size of a photon well is determined by two major parameters: The physical size of the image sensor, and the number of photon wells on the sensor. A large sensor obviously has more room for placing photon wells on its surface than a small sensor. And if we put few photon wells on the sensor, each can be larger than if we put a lot of photon wells there. In a Bayer sensor, one photon well roughly corresponds to one pixel in the processed image.
Sensors that are physically large can be manufactured with larger photon wells for a given number of mega-pixels. This is the reason DSLRs with large sensors, such as the Nikon D700 and Canon 5D mark II produces virtually noise free images at ISO 1600, while compact cameras with fingernail-sized sensors produces images of inferior quality at ISO 1600.
Given the same physical size of the image sensor, having fewer mega-pixels results in each photon well being physically larger. This is the reason the Nikon D3s (12 Mpx) produces cleaner images at high ISO than the Nikon D3x (24 Mpx). Both cameras have the same size sensor (FX), but the images from the D3s at ISO 1600 has about the same noise as the Nikon D3x at ISO 400 if we are looking at 100 % crops from both.
However, if we downsample the Nikon D3x image from 24 Mpx to 12 Mpx, to match the resolution of the D3s, the noise at the same ISO setting will be very similar. Downsampling reduces noise because it increases the amount of information (signal) that is used to determine the luminocity and colour values for each pixel.
While the physical size of the sensor and the total number of mega-pixels spread over that are the two major characteristics that determine the noise level of a digital sensor, there are other parameters that also influences noise levels.
For example: A sensor needs some supporting electronics to function. If these electronic components are placed on the sensor's surface, they take up real estate that reduces the area available to photon wells. Designs moving these components away from the sensor's surface, or making them smaller, makes it possible to increase photon well size and thereby also increasing the signal level. Placing micro-lenses over the photon well increases the area of the light-collecting surface without increasing the size of the photon well has the same effect (this is like placing a large funnel over a bucket to collect more rain drops). To some extent, more sophisticated signal processing may also improve the signal/noise ratio. Progress are still being made in the design of digital sensors and signal-processing, and we shall probably continue to see better noise characteristics as new image sensors become available.
Testing Image Quality
You may want to test your digital camera to see what kind of image quality it offers at different ISO settings.
Shoot the same scene at different ISO settings and see how high you can set the ISO speed and still get results of acceptable quality.
Many digital cameras have built-in high-ISO noise-reduction features that reduces high-ISO noise, but at the same time remove detail, leaving you with images that have an artificial, “plastic” appearance. If so, make tests with and without the noise reduction feature activated at different ISO settings and exposure times.
By learning the limitations of your camera before you need to push the envelope, you will save yourself from ending up with unusable photos from an important occasion because you picked a setting that produced low quality results.
Shooting at higher ISO settings increases image noise, shooting with noise reduction activated removes detail, shooting at longer exposure times increases the risk of motion blur. You need to find this out for yourself by testing all the possible combinations with your camera and see which results you prefer.
ISO Film & Digital Speed Documents
- ISO 6:1993
- Photography – Black-and-white pictorial still camera negative film/process systems – Determination of ISO speed.
- ISO 2240:2003
- Photography – Colour reversal camera films – Determination of ISO speed.
- ISO 5800:1987
- Photography – Colour negative films for still photography – Determination of ISO speed.
- ISO 12232:2006.
- Photography – Digital still cameras – Determination of exposure index, ISO speed ratings, standard output sensitivity, and recommended exposure index.
- Kodak application note MTD/PS-0234:2009.
- Kodak image sensors – ISO measurement.