A 2019 YouTube post by a well-known camera commentator provoked a lot of debate on the Internet, but it wasn't new debate. For quite some time the term ISO has been argued over. What exactly is it, and do we need to pay attention to it?
Some of the problem is this: the part of the sensor that converts photons into an electron charge—the photo diode—has a fixed response to light. If X number of photons get to the photo diode at a pixel location, then some number less than X—call it Y—will be converted into a charge that is then stored in the sensor.
Changing ISO does not alter that basic fact. For any given light condition, the charge stored for a photosite will be a fixed value, Y. What changing ISO does is alter what we do with the charge. And that's where most people start getting into trouble when talking about ISO for digital cameras.
ISO is a standard, though one that has some ability for those that use digital imaging sensors to game a bit. Funny thing is, the same thing was true of film, and we argued about ISO for film, too. Ansel Adams and his disciples were notorious for altered processing techniques to maximize the implicit light storage in the celluloid layers. Even slide films, such as Velvia, were shot by pros at a range of "ISO values" that, if I recall correctly, ranged by almost a full stop.
In both cases—film and digital—the thing exposed to light had a fixed amount of light hitting it. That light was in olden days converted to (eventually) free silver ions in most black and white film, while these days it’s converted to charge (electron storage) in digital. We alter the amount of light that gets to those surfaces by choosing apertures and shutter speeds, but once we've set those, the amount of light that gets to the focal plane is fixed for both film and digital (I'll have another article on that part soon). A single amount of light, not something that can be varied.
One big problem with ISO is this: what's the image?
That's not as strange a question as you might think. Kodak and Ansel Adams used to have fights over that. But even Kodak themselves weren't 100% into ISO. Some of their films had ISO ratings, some EI ratings, and some neither. Why? From a Kodak article (no longer available on the Web): "ISO speed ratings are determined under one set of very specific conditions. Sometimes those conditions, while highly standardized, do not correlate as closely as desired to real world conditions."
These days, digital camera makers generally make their "ISO decisions" based upon a finished image (e.g. JPEG, with all color and gamma corrections applied to form a final image; and that's almost certainly for the default JPEG rendering, not all the other possible JPEG renderings). In the film days, your processing and print decisions informed what you set your ISO to. So nothing's really changed. Only so many silver ions were created in our old black and white films, only so many electrons are stored in our digital cameras.
It's where we go from those ions or electrons that creates all the confusion with ISO.
So let's take the primary objections that have been arising in things like that YouTube post and address them:
- Camera makers don't comply with a standard. Actually, they do. The digital ISO standards all camera makers follow do have a couple of variations in how they can be applied, but every camera maker appears to follow the guidelines properly. For JPEGs. Not for raw data. From the standard guideline document: "ISO speed and ISO speed latitude values shall not be reported for raw images..." Uh-oh.
For raw files, we see clear differences in how camera makers place the data in the bit container, among other things. Some are trying to preserve headroom, some aren't. There's variation among those trying to preserve headroom as to how much headroom. Starting with the D5 generation, for example, Nikon made a change to their decision there, setting raw values so that they had more highlight headroom. Likewise, back around the D4 generation Nikon made a change about where black was placed in the bit container, which changed how the shadows behave. Also note that the camera makers don't know how color and tonal values will be resolved from raw data (ever notice that not all raw converters are alike? ;~) - High ISO values are just multiplication. Uh, yes and no. We actually have a fair amount of variation here among sensors and cameras. Many older and specialized sensor designs are not what we call ISO invariant (a linear multiplication). For those sensors—the Nikon D6 is one up through about ISO 800—you can't make this assumption of linearity. Ditto many Canon sensors. For most recent Sony sensors, they now use the Aptina-invented dual-gain variation, which means they're ISO invariant to a certain value, then change in the way they interpret the data and generate values, but are again ISO invariant above that setting. The Nikon D780 and Z6 are dual gain with the change at ISO 800, the Nikon D850 and Z7 do it a stop lower, at ISO 400.
What goes unsaid in almost every discussion of ISO values and whether they're needed or not—the premise of the video that's making the rounds at the moment—is what I call the tyranny of bits. And a related problem: integer-based math creating number rounding. I deal with this in my books, and for anyone shooting raw, it's something you need to understand.
Let's assume a truly bit-constrained digital imaging system for a moment. I'm going to use 4 bits here to show you what happens:
- 0000 = black
- 1000 = mid-tone
- 1111 = white
We can record 16 values with this system, and let's assume that we have 8 stops of DR in the sensor. What happens when we want to bring the darker values—say a value of 0011—up by a third-of-a-stop? Well, values are going to get rounded (and the 0011 data point was probably already rounded). We don't have enough bits to accurately make the change, so we either are rounding individual bits up or rounding them down (and worse still, some systems and software don't round correctly at all, they truncate).
Here's the dirty little secret you don't know: in order to make all your "edits" lightning quick, your raw converters and image editors aren't using precision math. Those of you with technical backgrounds will immediately recognize that there's a difference between doing all your math in 16-bit integer values versus doing it using 32-bit floating point values. The former is going to create rounding errors more often than the latter. Hey, and isn't your CPU 64-bit? ;~)
I once had a statistics professor try to ding me for using my own software instead of the commonly used SPSS of the time, because my final values were often slightly different. Yes, they were different: they were more accurate because my math was more precise! Unfortunately, most of your raw converters are using 16-bit integer math for speed because you've told the software developers you don't like waiting for anything, ever. And that means they're not exactly accurate.
So here's the problem with shooting only at base ISO and then post-processing your raw data to some other implied ISO (e.g. exposure boost in data conversion): you don't always have enough bits to be fully accurate (particularly in the shadows) and the math that's being used to make things happen fast isn't accurate, either.
Where people "fall down" in their ISO arguments is that they confuse subjective first impressions—it looks the same to me—with analytical and measured examinations that are accurate to the actual data.
Here's something to consider: it looks the same subjectively after you move the Exposure bar in Lightroom over a few stops, but now you apply another effect like sharpening, which moves pixel values some more. Are you absolutely certain that you're not just abusing the data at this point and not picking up unwanted artifacts? (hint: you shouldn't be certain; and did you know that most converters are already applying an unseen exposure correction they don't report? Thus, your belief that you've moved the values from a particular exposure to another aren't even right to begin with.)
There's long been what I consider to be a near gold standard in raw conversion, a Mac-only program by Andrey Tverdokhleb called Raw Photo Processor (RPP) (though it hasn’t been updated recently). He and my friend Iliah Borg (of Rawdigger fame) collaborated to put together a raw conversion tool that, frankly, can bring out more of what your sensor is actually doing than the Adobe converters can do (warning: RPP is geeky and complex, as is Rawdigger). One simple thing is that RPP uses more precise (and slower) 32-bit floating point math, but another very important thing is that it treats the two greens as different (uh-oh, you mean green on the red row is different than green on the blue row in a Bayer sensor?). RawTherapee is another converter that uses precise floating point math.
Remember the "it looks the same to me" trap I mentioned? Guess what happens when you use more precise math with more care in dealing with small, but real, differences? Yeah, you can bring very small things into very real visibility. And guess what? That's where the ISO invariance proclamations begin to completely fall down: as it turns out, the camera makers know what they're doing. The math in the ISP chips that is creating the raw data is somewhat more accurate than the math your *cough* Adobe converter is using. Not completely accurate, unfortunately. I mentioned Iliah. He's found several mistaken algorithms within those ISPs along the way. The whole Z7 banding issue is one of them.
Still, the most accurate data you have to work with in raw files is that which you'll get if you optimized exposure settings, and that means also using a "correct" ISO value rather than base.
Note these words in that last paragraph: "raw files". If you're shooting JPEG there's absolutely no way you should be shooting at base ISO and then varying "exposure" after the fact. The tyranny of bits will bite your butt big time if you try that. You get away with it in raw converters because they're "good enough" for those not paying close attention the details.
This is all just one reason why I don't buy into the "ISO is fake" articles (there's another, but I'll save that for another day, as it gets into the technical underpinnings of the sensor itself).
Finally, a comment that's a bit self-serving. The information you find on the Internet (about anything) tends to come in all sorts of forms and levels. At the bottom there's just outright misinformation in hopes of changing your mind/habit/buying. Just above that there's gross oversimplification, which can easily be misinterpreted. At the top there's peer reviewed, accurate, detailed, and testable facts and observations. I personally aspire to being closer to the latter, and well away from the former. Over time, I think the cream of information providers rises to the top. You should seek those out if you really want to know what's going on.
This article was triggered by many multiple emails asking "is it true?" Good on you to raise the question, it shows you're trying to figure out which sources of information can be trusted.
Thanks to Iliah Borg for reviewing a draft of this article.
