Behind the scenes: extended highlights!
As we saw in an earlier article, the ISO definition used when you take photographs is based on the brightness of the resulting JPEGs, not on any inherent property of the sensor. Which is something to remember whenever you see anyone talking about manufacturers ‘cheating’ on their ISO figures.
So long as the final JPEG image comes out at the correct brightness (which we'll be testing), manufacturers can do whatever they want with their sensors (and image processing). And, as we saw, if they’re using the Relative Exposure Index definition for pattern light metering, they get to choose what the ‘correct’ brightness is, as well. This provides the scope for manufacturers to do some interesting things to boost dynamic range.
When we reviewed the Olympus E-620, we noticed that the dynamic range was much lower at ISO 100 than at 200 and above. What was interesting about this was that our dynamic range graphs (that essentially plot the camera’s tone curve), followed the same pattern as those of the Nikon D5000.
This prompted us to look into what the camera was doing. We did this by shooting our standard dynamic range test at ISO 100, ISO 200 and ISO 100 underexposed by 1EV (which meant using the same exposures as the ISO 200 shots). Of course the JPEG of the underexposed ISO 100 shot was too dark but interestingly it clipped at exactly the same point as the ISO 200 JPEG.
|As can be seen, the correctly exposed ISO 100 JPEG produces significantly less dynamic range above middle grey, with the top of the tone-curve clipping to white much more abruptly. However, if underexposed by 1EV, ISO 100 clips to white at the same point as ISO 200 (albeit while producing a darker image).|
We then decided to look at the RAW values of the files, to see what was going on. This required the use of a rather esoteric RAW analysis tool (one that makes no attempt to do anything so prosaic as produce pictures) and found that, as we expected, the underexposed ISO 100 file was identical to the correctly exposed ISO 200 file (metadata aside).
|If you graph the scene brightness against the RAW levels being used to capture them, you can see that the ISO 200 and underexposed ISO 100 shots are using the same, lower raw levels to capture any given brightness. The manufacturer is simply choosing to use a different region of the sensor's response to capture the scene.|
Which presents an interesting conundrum: is the Olympus underexposing ISO 100 to produce its ISO 200 or is it overexposing ISO 200 to give an ISO 100 setting (Which is what Nikon implies about the D5000)? The answer is, of course, that there is absolutely no difference between the two statements. Or, to be absolutely precise, neither statement is correct because nothing is being overexposed or underexposed.
Simply put, the sensor is being used the same way at both ISO settings. The exposures differ by 1 EV and so utilise different parts of the sensor’s response, but different tone curves are then used for the two different ISO settings so that you get the ‘correct’ brightness, either way.
This technique may be different from the conventional method, which amplified the sensor’s output by different amounts at each ISO setting, with this amplification compensating for the lower exposure such that each ISO used the same region of sensor response. However, there’s nothing in the ISO definitions used by camera makers that dictates how the sensor must be used. Indeed, this technique of using a lower region of the sensor’s response isn’t at all uncommon – we could just as easily have looked at a Canon using Highlight Tone Priority or a Pentax in its DRange modes, they’re doing much the same thing (The ISO 100 disappearing as an option is the biggest giveaway).
|What happens when I change the ISO?|
|Traditionally ISO has been changed by amplifying the sensor's output before it is converted to digital data (as demonstrated if you move your mouse over the above diagram). However, it is also possible to mathematically manipulate the data once it has been digitised - many 'extended ISO' settings and some intermediate ISO values between full stops (e.g. 250 and 320) do just that.|
A little bit like the automatic distortion correction in the Micro Four Thirds system, you have to go to quite some lengths to see what’s going on behind the scenes. And, since all the most common RAW converters, including the bundled software, Adobe Camera Raw and Capture One honour or, at least, acknowledge the manufacturer’s rendering intent, it’s not clear why it would bother you.
If you’re shooting RAW and you want to choose which region of the sensor’s response you want to use, you might want to work out which tone curve is closer to the one you plan to apply, then use that ISO setting, so that the histogram and blinking highlights to tell you something useful about exposure, but beyond that, the usefulness of this knowledge is essentially academic.