What is ISO on a digital camera?

This post is derived from my web article: http://www.clarkvision.com/articles/iso

What is ISO on a digital camera?

Reading manufacturer literature, review sites, and online discussions, one is led to believe that ISO changes sensitivity, and changes exposure. These ideas are both incorrect.

Myth #1: ISO changes sensitivity. False! Digital cameras have only one sensitivity, given by the quantum efficiency of the sensor, and the transmission of the optics and filters over the sensor. ISO is simply a post-sensor gain applied to the signal from the sensor.

Myth #2: ISO changes exposure. False! When you change ISO on a digital camera, two things are changed if the camera is in one of several auto modes. ISO changes only one thing in manual mode. In an auto mode, either exposure time, or f/ratio, or both are changed. F/ratio changes the diameter of the iris diaphragm which controls how much light gets through the lens. Exposure time limits the time light is collected by the sensor. It is the lens aperture and exposure time that determines exposure, not ISO. The second thing that changes with ISO, whether in manual mode or an auto mode is the range of light that is digitized.

A Savings Account Analogy

Say we have a bank savings account and we want to know how much money is in the account. Say one can count only 100 units. If I choose dollars, I can count from 1 to 100. If I use 10 dollar bills, I can count by tens from 10 to 1000. I can count in pennies, but then my range is only $0.01 to $1. Which level one chooses to count will be dictated by the amount of money in the bank account if one wants to know the full amount of money in the account. If we have $623.47 in the account, we must count by tens, and we would determine we have $620 in the account. We could count by twenties, and still find we have $620. If we count by fifties, we would find $600. We can never get a precise amount in the account, but we can get a close approximation.

Digital Cameras

So too with digitizing light recorded by a digital camera: ISO chooses a range to digitize. A given lens delivers so much light per second to a pixel in a given scene, and the exposure time sets the total amount of light that could be recorded. We choose the ISO as a signal range in which to digitize the light the sensor captured. Low ISO is likechoosing counting by tens with the bank account. High ISO is like counting by pennies in the bank account. Changing ISO is only changing the way we count, not how much light the sensor actually captured. If we choose too low of an ISO, we can't count high signals and we say the image is saturated.
So why not set the ISO low to capture the full signal range? Electronics in current cameras, especially DSLRs which have sensors that can capture light over a very large range, are not capable of digitizing the full range of signals. Further, some camera electronics add fixed pattern noise into the image and the human eye-brain finds such noise more objectionable than random noise.

But you may say when I change ISO in my camera, I record different amounts of light (in an auto mode). Let's sayyou are working in aperture priority mode where you choose the aperture and the camera meters the scene and choose the exposure. Let's say you are at f/5.6 and 1/500 second at ISO 100. You change ISO to 200 and the camera changes the shutter speed to 1/1000 second. Thus, the sensor now records half the light. So the ISO change instructed the camera to shorten the exposure time, thus recording less light, and then instructed the camera electronics to amplify the signal from the sensor and digitize a smaller range. This process means that any noise will be amplified too, and that is why we see more noise at higher ISOs. So the bottom line is the higher ISO decreases true exposure, recording less light, and amplified the smaller signal to compensate.

The top digital cameras have an incredible range of light that can be recorded. For example, the Canon 1D Mark IV can digitize up to about 55,000 photons at ISO 100, but the sensor read noise is only 1.7 electrons, making a dynamic range of 55000/1.7, or over 32,000 (15 photographic stops). A 14-bit analog-to-digital converter (A/D) can only digitize a range up to 16384 (14 stops) but the noise in the A/D and other electronics in many DSLRs limits the effective range to less than 4096 or less than 12 stops. This means we must choose ISO to digitize the range of light we want to record. To digitize a lot of light we use low ISO (low gain), and use longer exposure times or larger apertures to gather more light, but we lose the fine intensity (tonality) details, similar to not being able to count pennies in our bank account analogy when we had a lot of money to count.

To add further complications, camera manufacturers define ISO differently for each camera. It is a relative level of light compared to the maximum signal a pixel is capable of recording. Smaller pixels generally have lower holding capacities, and it goes roughly by the area of the pixel. So a pixel with half the area, will generally have a maximum holding level half that of the larger pixel. An analogy is two buckets of different sizes: the larger bucket can hold more water. But the larger bucket only holds the water poured into it. With cameras, the lens is "pouring" the light onto the sensor. The lens and exposure time determines true exposure, not the size of a pixel nor ISO.
Let's examine some real capacities. The data are from my digital sensor performance web pages at:
http://www.clarkvision.com/articles/index.html#part_4


See the full Table 1 at: http://www.clarkvision.com/articles/iso/

(I couldn't get it to format correctly here on BPN.)
The bottom line for the discussion about Table 1 here is that the maximum signals for the 7D, 1DIV and 5DII are:

__________ 7D ____1DIV___5DII

ISO 100: 24800 _ 55600 _ 59400 (electrons)

ISO 200: 16300 _ 27800 _ 29700 (electrons)

ISO 400: _8150 _ 13900 _ 14800 (electrons)

So let's say you have lenses that on each of the above 3 cameras that deliver the same photons per second to a pixel. Then if we set the exposure time the same on all three, each camera gets the same true exposure. Let's say that the lenses and exposure time deliver 30,000 photons to the sensor and each sensor captures 40% of the light, so gets 12000 photons in each pixel. The ISO needed to record that range (0 to 12,000 photons) is:

7D: ISO 200, 1DIV: ISO 400, and 5DII: ISO 400.

Note that even though the ISO is set differently on the 7D, it is still digitizing very close to the same signal range of light, and regardless of the ISO setting, the true exposure is the same.

In Table 1, Canon could have set the gains in the electronics such that all 3 cameras recorded the same signal range at the same ISO. That would mean, however, that the 7D would not have an ISO 100, because the pixels are not large enough to hold more than about 25,000 electrons. It would mean, however, that when people compared results from the cameras, and used the same ISO, they would be comparing the same true exposure. However, since the gains are not equal, most comparisons one sees on the internet use the same ISO, which biases against the 7D as it gets less true exposure and comes out noisy, but such conclusions are in error and a result of unequal exposure.

Conclusions

ISO is a relative gain, varying by camera and has nothing to do with sensitivity or true exposure. ISO is simply a post sensor gain and digitization range.

Future Prediction

Future cameras will have 16-bit A/D converters thus reducing the need for ISO gain. Actually 18-bit A/D converters are needed for the top cameras, but we are not likely to see that anytime soon. With 16-bit A/Ds, we'll only need broad categories of ISO, like low, medium and high, and choose precise gain (ISO) with raw conversion.

Roger