Living with Photography: ISO Don't Get It

By Norman Chan

ISO is one of the three pillars of exposure--usually set after aperture and shutter speed, and in digital cameras, is actually a manufactured analogue from film camera technology. Yes, the ISO setting in your DSLR is a skeuomorphism. Let's dive into how it works and how I'm learning to use it.

Last week's light dive into the technical workings of lens aperture and the F-Stop was a ton of fun; I really enjoyed doing the research into a topic I only understood at a surface level, and then talking it over with you guys. So this week, I've been doing the same thing with another technical aspect of photography, camera ISO. ISO is one of the three pillars of exposure--usually adjusted behind aperture and shutter speed, and in digital cameras, is actually a manufactured analogue from film camera technology. Yes, the ISO setting in your DSLR is a skeuomorphism.

But before we dive into technical details--seriously procrastinating that for as long as possible in this column as I digest the information myself--I want to talk about how I've been using ISO for the past year, and what's changed recently. If you follow my camera adventures, you know that since getting a mirrorless APS-C camera in 2011, I've been addicted to shooting with low ISO to maximize potential image quality. Maybe it was the mental trauma from years of using a Canon 30D, which produced grotesque noisy photos at ISO 800 and above. Low-light photography without the use of a flash on that camera was really difficult, compared to the pristine photos I was getting on the NEX-C3. Since then, against the recommendation of smarter people, I've persisted with a low-ISO-only attitude.

But this year's CES was when I finally wised up. Based on the recommendation of one of you in the comments, I shot all of my CES photos using my camera's auto ISO function. Mind you, I only gave into doing this because the Canon 6D has a neat feature that lets me set both an upper limit to auto-ISO and a minimum shutter speed to let the camera to raise that ISO. And even though a configuration of ISO 3200 max and 1/125s min was suggested, I still wussed out and put my auto ISO settings at ISO 1600 max and 1/80s min. That means that the camera would do its best to keep the shutter above 1/80s for every photo (serviceable for slow moving scenes), bumping the ISO to at most 1600 for my given aperture and desired exposure. My goal was to not have to think about ISO at all throughout the week, and analyze the chosen settings for each photo after.

Running through the EXIF data on my 50 favorite CES photos, I was pleasantly surprised at the distribution of auto-selected ISO settings. While I don't have this data plotted out on a graph, a majority of the photos were set to ISO 500 or lower, with a few evening event photos bumped to 1000 and only a handful at the max 1600. Convention hall and booth lighting is actually relatively bright and conducive to photography; I would mostly adjust desired exposure to compensate for poor metering in reflective shots, or to intentionally blow out backgrounds to highlight a product. Auto ISO at CES was a definite success.

ISO 200 (left), ISO 1000 (right)

But last week, as I was going through my evening ritual of Lightroom processing after a day shoot with Adam at the Cave, I found myself being very disappointed with the look of the photos I had taken that day. For example, in the photo comparison above, the one on the left was my favorite photo I took of Adam last year, at ISO 200, f/2.8, and 1/100s. The one on the right, also taken at f/2.8, was automatically set to ISO 1000 and shutter of 1/80s. When zoomed in to 100% scale (click the image for that), I was really unhappy with the graininess of the details on Adam's face, notably in his glasses. Is the photo "usable?" Absolutely. But it's pales in comparison to the clarity and definition of the photo on the left. Sure, the lighting situations were completely different between the two, with the natural light on the left being much more ideal. But that's something I never took for granted when limiting myself to low ISO--I would seek out good light to be able to shoot at ISO 100-400. With Auto ISO, I'm taking the ability to shoot in lower light for granted, and as an effect am taking more shots with uninspired light placement. Auto ISO has the risk of becoming a crutch if you don't pay close attention. That's personal lesson one.

OK, so even with Auto ISO, you have to be mindful of light placement. But what about the limits of raising ISO for the usability of a photo? Every camera make and model implements ISO settings differently (again, putting off the technical talk until the end), and I realized that I never did any testing of my Canon 6D's ISO settings. DPReview's comprehensive ISO noise and detail tests are thorough enough for most people, but it's still fun to run some anecdotal photo comparisons yourself. For the two following sets of photos, I assembled simple setups in a relatively dark part of the office, using nearby room lighting and placing my camera on a tripod. Noise reduction was set to default and I shot in RAW, exporting the full-resolution JPEGs from Lightroom afterward. The lens was my 28-70mm set to f/2.8, and exposure compensation was set to 0. A proper test would use a much lower aperture for wider depth of field and higher potential sharpness, but I wanted to see high ISOs' effects on bokeh as well as in-focus text and detail.

ISO comparison for Canon 6D, "Save Link" for full image.

In this first series of test shots, I placed the Hellboy Samaritan at an angle across the spine of the book. Focus was placed on the spine text, and a bright lunchbox placed in the background to study noise in bokeh. On the pistol itself, I tried to angle it in a way to get the grooves of the handle in focus, as well as part of the trigger guard and barrel. The angled barrel would let me see how noise affects the transition of focus with a shallow depth-of-field.

And the result? After scrutinizing over the details of my nine shots--from ISO 100 to 25600--I noticed interesting trends on the low, middle, and high end of the ISO settings. On the low end, I realized that it was actually pretty difficult to distinguish between photos in the ISO 100 to 400 range, especially when exported as JPEGs with minor compression. The places where I looked for differences the most were in the in-focus text on the spine and the shadows, like the white book pages between the pistol grip and the book's spine. Up to 400, noise in that shadow area was hard to see.

The increase in noise between ISO 800, 1600, and 3200 was more distinguishable, but only in a sense that I could easily tell between the three photos. At ISO 1600, the photo still looked very clean (admittedly shot at 1/6s), and the noise was more a reduction in clarity than in sharpness or color accuracy. At ISO 3200, sharpness in the details was still good, but I could see the noise creeping on color fidelity. The green of the book spine, for example, was beginning to look a little greyed out.

At ISO 6400, 12800, and 25600, however, noise was becoming not only very noticeable, but the amount of noise introduced with every stop was dramatic. Detail in the out-of-focus areas was smearing to a point where you couldn't get an accurate sense of depth with the bokeh. The top edge of the pistol barrel, for example, looked just as far away as the image on the lunchbox. The good news was that when resized for web--for example to 800 pixels wide--the effects of downscaling mitigate a lot of the noise. I would go so far as to say that even ISO 12800 images can look good when scaled down. ISO 25600 gives the look of poor compression.

ISO 200 vs 1600 vs 6400 on Canon 6D

The next series of tests was similar to the first, and my conclusions about the same. An interesting thing I noticed was that I actually preferred the quality of detail at ISO 200 moreso than at ISO 100, when looking at the edges of the frame. The fibers of the Wilson tennis ball on the right, for example, looked better for some reason at ISO 200 than 100. And as with the first test, ISO 1600 was a sort of sweet spot for acceptable noise at full resolution without noticeable color and dynamic range degradation. And once again, ISO 12800 was surprisingly usable once the photo was downsampled.

ISO comparison for Canon 6D, "Save Link" for full image.
ISO 200 vs 1600 vs 6400 on Canon 6D

OK, let's actually get technical and talk about how ISO works and why it differs from camera to camera (as best as I understand it so far).

ISO in film cameras was commonly known as the film's "speed," though it had no bearing on the actual speed in which the film moved through the camera. This was akin to talking about the speed of a lens. Faster or higher ISO film was a reference to the chemical composition of the film--the quality of the light-sensitive silver halide chemicals used on the film. Silver halide is put on film in crystalline form, and it's the density and size of these particles that determine the film's ability to capture light. The larger the particles, the more photons can be captured on the silver halide, meaning that the shutter speed or aperture can be reduced to get the same exposure as film with smaller particles. But larger particles also correlates with the lower density of these crystals, and the high granularity of silver halide crystals in high ISO film is what causes the optical effect of film grain. Changing the temperature or time of developing of the film can also affect the film grain; the point is that it's both a chemical and optical process.

In digital cameras, there is no chemical process. And if the size of a frame of film is analogous to the size a digital sensor, the photo-sensitive silver halide crystals in film are analogous to the individual pixels on the camera sensor. The important difference, though, is that digital sensors can't adjust the size of their pixels on a whim, as you could change out the film stock in a film camera. A 20MP sensor has 20 million pixels of a fixed size--each a little bucket for receiving light.

How much light each of these pixel "buckets" receives until they're considered full, then, is what determines the ISO of a digital camera. Raising the ISO on a DSLR tells the sensor that for each of its pixels, it only needs to store a fraction of its capacity for photons (stored as an electrical charge). For each of the these tiny pixel wells, the stored charge is run through an Analog-to-Digital converter to translate the accumulated light into a RAW digital file. Because at higher ISOs, the charges are a fraction of what they would be for a normal exposure, the charge is amplified to increase the gain in electrical signal before going through the A/D conversion. That voltage gain--an analog process--is what creates digital noise in the image data. It's an electrical anomaly, as opposed to an optical or chemical one.

So the reason different cameras produce different quality photos at the same ISO settings is because camera makers have different technologies for putting a gain on its pixel charge, converting that charge to a digital signal, and also reducing that noise with built-in computational algorithms.

There are a few takeaways from understanding how ISO works in digital cameras. One is that digital sensors have a "native" ISO for which its pixels store their maximum charge capacity. You would think that this is ISO 100, but it turns out that it can vary between models. Nikon used to publish the native ISOs of its cameras, but Canon never has--hardcore camera users have deduced native ISOs through testing and reported their findings on various forums. Setting a sensor to below its native ISO--what happens when a camera has an ISO of 50, for example--doesn't eliminate noise; it actually reduces the dynamic range of the photo because the signal is attenuated in the A/D conversion (the opposite of amplification). Conversely, when cameras let you push sensor ISO to beyond the "normal" range (eg. up to 102400), it's actually just scaling up the image data in processor. You would get the same or better effect scaling in post-processing.

ISO, like shutter, also increases linearly in terms of light stops. Each doubling of the ISO setting doubles the potential exposure. But there are many more than nine ISO settings on my camera, between 100 and 25600. In those mid-stop settings, the camera sensor actually sets its pixels to the nearest ISO stop (eg. ISO 500 is actually ISO 400) and then overexposes or underexposes the image to match that "speed." There's some debate online about whether it's ever worth using those mid-stop ISO settings or to always stick to the even stops--100,200,400,800, etc. But that's beyond my comprehension of the topic and an exploration for another day.

For further reading, I recommend this piece on film and digital camera ISO, and this one on ISO gain/amplification technology.