What is ISO: A Technical ExplorationBy Rob Taylor,11 Mar 2013

Most people understand the practical use of ISO, but what is it, where does it come

from, and what's difference between ISO in film and digital? I'm going to explore

the history and technical underpinnings of the system. If you've ever wondered

what ISO means or how it works, this one's for you!

The History

ISO, in its photographic context, is the standard rating system of the light sensitivity

of a photographic medium. It's the acronym for the International Organisation for

Standardisation, a global body who work to standardize all kinds of products and

processes for maximum interoperability and safety.

They codified the ISO film ratings in 1974, combining the most recent advances in

the German DIN and American ASA (now ANSI) systems into a single universal

standard.

These two systems stretched back to the 1930s and 40s, before which various

ratings systems coexisted from different manufacturers and engineers, despite

35mm film being accepted as the international standard back in 1909. 120 medium

format film also dates from around this time, but the large film size increased its

cost and therefore reduced its overall popularity with amateurs.

How It Was Measured

What do the numbers themselves mean? There are four ISO standards, which

govern color negative film, black and white negative film, colour reversal (slide) film

and digital sensors. These are calibrated so that regardless of the type of film or

medium, the effective sensitivity is theoretically the same.

This is useful for practical mathematical purposes while shooting, although

photographers have usually found that for some films, setting cameras to slightly

different ISO ratings than a particular film's nominal speed gives better results.

The differences in emulsion and interpretations of measurement processes across

manufacturers, factories and even batches, as well as the inherent variability of a

chemical process, means that even with standardisation, results can vary.

In recent times, film speed has been measured from a "characteristic curve," which

describes a film's general tonal performance. This curve is created using a

"sensitometric tablet," a sort of graduated ND filter consisting of a precisely

calibrated array of 21 equally-spaced (from black to white) shades of grey.

They are exposed onto the film in a sensitometer - a light, shutter, filter holder and

film holder. After processing, this results in a stepped graduation in the optical

density (ie. darkness and/or opacity) of the emulsion on the exposed section of

film.

The 21 steps are then each measured using a highly accurate instrument called a

densitometer, which shines a light through the film at a photodetector and gives a

reading on a scale of zero to three. Once all 21 steps have been measured, they

are plotted on a graph in millilux-seconds.

This graph has various parts which explain various aspects of the film such as

fogging, gamma, contrast, etc. The part we're interested in for the ISO speed rating

of the film is 0.1 density units above the minimum density, let's call this point x.

This value isn't particularly scientific, but is traditionally accepted as the minimum

difference in density that the average human eye can differentiate.

The equation for film speed (yes, there is one) is $$speed = {800\over{log^{-1} (x)}}

$$ If the exposure is measured in lux-seconds rather than millilux-seconds, this

becomes: $$speed = {0.8\over{log^{-1} (x)}}$$ Note that I write log for base-10,

not ln for natural log (base-e). As the speed doubles or halves, so too does the

sensitivity to light.

How the Sensitivity ChangesFilm is made of a suspension of silver halide crystals in a gelatin binder. This

emulsion is finely layered many times along with any dyes for color or processing

agents onto a celluloid base, protected on the back side with physical handling

coatings. The silver halide crystals are the actual photoreactive medium.

They are only reactive to the blue end of the visible light spectrum (hence the need

for UV filters when shooting film), they're coated or impregnated during growth-

with organic compounds which sensitize them to the full visible spectrum.

Photons hitting the silver halide or the spectral sensitizers impart their energy into

the molecule. This causes an electron to be ejected from a halide ion in the silver

halide crystal. This can be trapped by a silver ion to form an electrically neutral

silver atom.

This is not stable, however. More photoelectrons must be available in the same

region to form more silver atoms in order for a stable cluster of at least three or four

silver atoms to be formed. Otherwise, they can easily decompose back into silver

ions and free electrons. More silver atoms can form as long as photoelectrons are

being generated.

An atom cluster of pure silver of this stable size will catalyse the reaction with the

developer, which then decomposes the whole crystal into a metallic silver grain,

which appears black due to its size and unpolished surface.

The fixer then fixes the image by dissolving the remaining silver halide salt crystals

which are then rinsed away. This has been the general basis of photography for

over a century. So what does this have to do with the sensitivity of film?

The answer to that is really quite simple: probability. The larger the silver halide

crystals, the more likely it is that photons will hit them and be absorbed. To use a

basic analogy, if you wave a large butterfly net through a large swarm of butterflies,

you're likely to catch more of them than with the same wave through the same

swarm with a small net.

Larger crystals have a greater surface area facing the lens, and logically, light

sensitivity directly correlates with the likelihood of light hitting the surface.

Thus slow films like ISO 25, 50 and 100 have very fine grains to reduce the amount

of light hitting them, useful for capturing fine detail. Conversely, very fast films like

ISO 1600 and 3200 have relatively huge grains for the maximum possible chance

of capturing photons, hence their extremely grainy quality.

How It Works for DigitalDigital cameras, having no chemical process, cannot be measured using the same

method as film. The ISO ratings system, however, is designed to be reasonably

similar to film in terms of actual light sensitivity. Technically the term for digital

sensors is "Exposure Index" rather than "ISO," but because an ISO standard

covers it, I see no issue using the more traditional "ISO."

Instead of a minimum visible exposure level, digital sensors have their sensitivity

determined by the exposure required to produce a predetermined characteristic

signal output. The ISO standard governing sensor sensitivity, ISO 12232:2006,

relates five possible methods to determine sensor speed, although only two of

them are regularly used.

A camera's sensor consists of a matrix of millions of microscopic photodiodes,

usually covered with microlenses for extra light-gathering and a Bayer pattern filter

in order to capture color. Each one represents a single pixel.

A photodiode can be run in either zero-bias (no applied voltage) photovoltaic

mode, where output current is restricted and internal capacitance is maximised,

resulting in a photoelectron build-up on the output.

It can also be run in reverse-biased (run backwards) photoconductive mode, where

photons absorbed into the p-n junction release a photoelectron that directly

contributes to the current flowing through the diode.

Camera sensors use the latter, as the voltage applied to reverse bias the diode

both increases the ability to collect photons by widening the depletion region and

reduces the likelihood of recombination due to the increased electric field strength

pulling the charge carriers apart. Suddenly lost? Let's go over the operation of the

photodiodes that make up the sensor in your camera.

A (Somewhat) Basic Interlude on PhotodiodesA photodiode is essentially a normal semiconductor diode (a device which allows

the flow of current in only one direction) with the p-n junction exposed to light. This

allows photoelectrons to have an impact on the electronic operation of the device.

A p-n junction is a piece of positively-doped semiconductor fused with a piece of

negatively-doped semiconductor. Doping is infusing impurities which donate or

accept electrons in order to alter the availability and polarity of charge in a piece of

semiconductor. This selective manipulation of charge is the basis of all electronics.

Close to the junction point in the semiconductor, the electrons on the negative-

doped side are attracted to, and tend to diffuse into, the postive-doped side. There

are holes without electrons within the semiconductor lattice, resulting in a net

positive charge. Holes are treated as positively-charged particles for general

purposes. These equally have a tendency to diffuse into the negative-doped side.

However, once enough mobile charge carriers (the electrons and holes) have

accumulated in each side, there is enough charge there to generate an electric

field which tends to repel more charge carriers from diffusing. A charge equilibrium

is reached. The diffusing carriers are equal to the repelled carriers in each

direction.

This equilibrated area near the junction is what's called a depletion region, where

there's a cloud of electrons on the positive-doped side of the junction, and a cloud

of holes on the negative-doped side. The carriers have been depleted from their

original positions, and have created a charge difference, resulting in an electric

field, ie. built-in voltage potential. This is the basis for a diode. A photodiode is

essentially the same thing, but with a transparent window to allow photons to hit

the depletion region.

Reverse-biasing the diode widens the depletion region by overcoming the natural

charge equilibrium of the depletion region and setting a new one, where the innate

electric field must now be strong enough to oppose both the attraction diffusion and

also the applied electric field. This, of course, requires a larger depletion region

containing more charge to generate a stronger field.

When a photon of sufficient energy hits and gets absorbed by the semiconductor

lattice, it generates an electron-hole pair. An electron gains enough energy to

escape the atomic bonding of the lattice and leaves behind a hole. Recombination

can occur immediately, but largely what happens is that the electron gets pulled in

the direction of the negative-doped region and the hole towards the postive-doped

region.

Often they can recombine with other charge carriers in the semiconductor, but

ideally, with optimised transit distance from the photosite to the electrode collector

(short enough to avoid recombination, but long enough to maximise photon

absorption) the carriers will reach the electrode and contribute to the photocurrent

to the read-out circuit.

The more photons are absorbed, the more charge carriers make it to the

electrodes, and the higher the current read-out sent to the A-D converter. The

higher the current, the higher the exposure being received and the brighter the

pixel.

How This Affects ISO

As I mentioned above, ISO is often measured using the exposure required to

saturate the photosites. I just explained what the photosites are; the depletion

region within the photodiodes. So how do they become saturated? Well, the

number of electrons available for photons to excite is not unlimited. After a certain

amount of light energy is absorbed, the semiconductor has released as much

charge to the electrodes as it can, and no longer responds to further exposure.

Photographically, this is the full-well capacity, or highlight clipping point. Usually

manufacturers deliberately mis-rate their sensors in order to retain headroom in the

highlights, allowing highlight recovery in RAW.

According to ISO 12232, the equation to define saturation-based speed is $

$S_{sat} = {78\over{H_{sat}}}$$ where $$H_{sat} = L_{sat} t$$ [latex]L_{sat}[/latex]

is the required illuminance for a given exposure time to reach sensor saturation.

The 78 is chosen such that an 18% grey surface will appear exactly 12.7% white.

This allows highlight headroom in the final rating for specular highlights to roll off

naturally and not as blocky dots. This rating is most useful for studio photography

where illumination is controlled and maximum information is required.

It defines another rating test which is lesser-used but is more useful for real-world

scenarios, which is the noise-based speed test. This is a rather subjective test, as

the image quality and test criteria are somewhat arbitrary; the signal-to-noise (S/N)

ratios used are 40:1 for "excellent" IQ and 10:1 for "acceptable" IQ, based upon

viewing a 180dpi print from 25cm away. The S/N ratio is defined as the standard

deviation of a weighted average of the luminance and chrominance values of

multiple individual pixels in the frame.

Standard deviation is a way of mathematically deriving the variation in values in

collected data from the average or expected value. It's the sum of all the

differences squared, divided by the number of data points in the set, square rooted.

Essentially, an average of the deviations.

Photographically, this means that the test pixels are averaged out to find the

"expected" value of the light signal. Then the standard deviation defines how far

away the individual test pixels tend to be from this average. Assuming the pixels

are relatively uniform in value, this deviation from the average is noise, either from

the sensor or the processing electronics.

The ratio between the average value (signal) and the standard deviation (noise) is

the S/N ratio. The higher this ratio, the less noise there is in the signal. For

example, for the "excellent" image quality standard of 40:1, this means that on

average, for every 40 bits of image signal, there's only one of noise. The huge

difference between the image and the noise is what creates the clean image.

Noise can be introduced in several ways: saturation/dark current across the

photodiodes, random thermally-released electrons in the photodiodes or

processing electronics (thermal noise), charge carrier movement across the

depletion region of the photodiodes (shot noise), and imperfections in crystal

structure or contaminants which result in random captures and releases of

electrons (flicker noise).

The increase in noise from increasing the ISO setting on the camera is a result of

increasing the gain of the pre-amplifiers between the sensor and A/D converter.

The S/N ratio is necessarily reduced, as in order to produce a "correct" exposure

with high amplification, there must be less exposure. Less exposure means less

signal, thus relatively greater noise as a fraction of that reduced level.

A simple mathematical example; say at ISO 100, a correct exposure is achieved by

filling a particular pixel to 80% well capacity, and its S/N ratio is 40:1, so +/-2% of

the current readout is noise-induced. Boosting the ISO to 800 means that the

amplifiers are boosting the signal by 8x, and thus the correct exposure is reached

at only 10% well capacity. The +/-2% noise level, however, remains about the

same and gets amplified right along with the signal level. Now that 40:1 S/N ratio

has become a 5:1 ratio, and the image is useless.

Conclusion

You can see why it's important to shoot with as much exposure and as little

amplification as possible. Circuitry and sensor technology, as well as denoising

algorithms, are constantly improving, just think about the difference between an

ISO 800 shot from 2008 vs an ISO 800 shot from today. The majority of images are

also now viewed at relatively small sizes online, and resizing also reduces noise.

For large format printing purposes, though, you can see why it's vital to shoot with

lots of light and at base ISO. Hence also the maxim "expose to the right," meaning

get the image as bright as possible on the histogram without clipping highlights.

Not only does that maximise the amount of light signal compared to the reasonably

fixed noise level of the imaging electronics, but the way the data is digitised means

that more information can be stored in the highlights than in the shadows.

That's about it, I think. I hope this article was of interest, possibly even use, to

some of you, and that you didn't get too lost in the technicalities of solid-state

physics!

Comments? Questions? Hit up the comments below!