Sensitivity (ISO) in digital imaging seems to be the subject of quite a lot of confusion - it’s becoming common to hear talk of manufacturers 'cheating with ISO.' So we thought it made sense look at why sensitivity appears hard to pin down, why we use the definition we do and how it’s actually not as complicated as it can sometimes seem.
ISO in Photography
Before we get too carried away with the intricacies of ISO standards, it makes sense to step back and consider how we use sensitivity in photography. Sensitivity is the connection between the physical exposure (how much light you let in) and the brightness of the final image. As such it joins shutter speed and aperture as one of the three factors that define exposure.
Sensors and sensitivity
For ISO to relate exposure to final image brightness, we have to think about the inherent sensitivity of the digital sensor and this is where it risks becoming rather removed from photographic concerns.
Much of the complication arises from the fact that there is no 'correct' way of exposing a sensor. Sensors have a capacity for converting light into electrical charge - limited at the upper end by the point at which the sensor becomes saturated (and cannot convert any more photons of light into electrical charge) and extending down until the signal is drowned-out by electrical noise. The upper, saturation limit of the sensor’s response defines the brightest light intensity that can be turned into meaningful data in the final image. However, this doesn’t tell us much about how to expose the sensor - simply exposing to retain the brightest highlights won’t necessarily ensure a correctly exposed image – we need to work out how to correctly expose a middle grey.
|
|
And there is an added complication before we can get to that point. Sensors respond to light in a very different way to the human visual system - they respond in a linear fashion: twice the amount of light gives twice the signal, whereas the brain doesn’t interpret things that way. In order to make this linear data into a convincing image, a tone curve that attempts to map the data back to the way the eyes respond has to be applied.
This tone curve converts the sensor’s output to the final image brightness, which means it also defines how the sensor needs to be exposed. (In fact there is a subtle interplay between the sensor’s inherent sensitivity, its dynamic range, the tone curve and the camera’s metering.)
A standard with shades of grey
So this is what ISO is defining when you use it: it’s combining considerations of the sensor’s sensitivity with the effects of the tone curve and metering so that you can get the correct final image brightness with your chosen exposure.
However, the standard set down by the International Organization for Standardization (ISO12232:2006, as it happens), contains five separate definitions, each of which can produce a different answer for the same camera. Thankfully, only three of these definitions are widely used and only two, closely-related definitions are used by camera makers.
ISO, courtesy of CIPA
The two definitions of ISO that are actually used by camera manufacturers (and are reported by their cameras) are based on the brightness of cameras’ JPEG output. Both definitions come from standards developed by the Japanese camera trade body CIPA, which were adopted by ISO in 2006. The first definition is probably the simplest and most intuitive, and it’s called Standard Output Specification. Essentially, it defines ISO as the camera behaviour that renders middle grey at the correct brightness (as we’ve just described and pretty much the same way as it did for film).
The other definition (Recommended Exposure Index) is fairly similar but is designed to accommodate multi-zone/pattern metering systems. These metering systems aren’t based on trying to represent middle grey and instead aim to achieve whatever the manufacturer considers to be ‘correct’ exposure. As such they can’t be measured because the definition is pretty much circular: whatever the camera chooses is right, by definition.
So what about the others?
The only other definition of ISO you’re ever likely to encounter is one that can be used for RAW data. The problem is that it’s based on a combination of the sensor’s saturation point and a generic tone curve – which isn’t necessarily the tone curve your camera’s JPEGs or metering are based on. So, discrepancies between this figure and your camera’s reported ISOs aren’t the result of under or over-reporting of ISO, they’re a measure of how different your camera’s tone curve is from this generic tone curve.
Why do I need to worry?
If you use the camera’s JPEGs, or a RAW converter that acknowledges the manufacturer’s rendering intent (and that includes many popular RAW converters), then chances are you’re going to get the ISO that your camera tells you. So rather than measuring a slightly obscure aspect of sensor performance, our tests are based on the Standard Output Specification that the camera manufacturers use, that your camera is based on and that, chances are, you use.
|
What happens when I change the ISO? |
REI is *Recommend* Exposure Index, not *Relative* according to the link you provided, and I quote:
"The Recommended Exposure Index (REI) technique, new in the 2006 version of the standard, allows the manufacturer to specify a camera model’s EI choices arbitrarily. The choices are based solely on the manufacturer’s opinion of what EI values produce well-exposed sRGB images at the various sensor sensitivity settings. This is the only technique available under the standard for output formats that are not in the sRGB color space. This is also the only technique available under the standard when multi-zone metering (also called pattern metering) is used."
Posted by: Terry Wyse | 23 Sep 2009 18:06:00
Sorry, typo on my part.
Posted by: Richard Butler | 23 Sep 2009 18:10:18
Wow, clear as mud. I thought I completely understood this and now I understand that I don't understand.
Does this explain why my RAW Canon exposures (5D) are usually a bit overexposed in ACR but perfect in the camera, because of the difference in the tone curve applied?
Is there no way to come up with an ISO standard based on a RAW capture that could be applied or is our best shot just a mid-grey reference?
What does DPR use as the basis for ISO, what method?
I can't wait for part II
Posted by: Kachadurian | 23 Sep 2009 18:44:18
But why should REI be a standard, if sensitivity can be vary from one camera model to the other and from one maker to the other, then what make it a standard? What stop a manufacturer from labeling ISO100 as ISO1600, and claim their new camera have 5 stops ISO advantage?
Posted by: Freeman Jotographer | 23 Sep 2009 19:08:35
Thats probably cause ACR and LightRoom have pre-defined settings for each camera (made by cooperation between Adobe and camera maker). For example, default for Panasonic G1 is +25 contrast +50 brightness in LightRoom (and other stuff, like chroma reduction).
Unless you turn this feature off, it will automaticly correct every imported picture..
Difference between what you see on your in camera LCD can be caused by various reasons.. starting different type of LCD used for camera and for your monitor, going over wrong default setting in ACR, or just badly balanced monitor?
Easiest way would be to correct automatic in ACR to underexpose slightly or expose properly..
Little bit wierd is that 5D has measured almost nominal ISO (tiny bit lower than nominal). So unless camera LCD shows image with some exposure correction, its probably ACR fault.. Difference won´t be tone curve but just exposure/brightness/contras setting..
If I´m correct, DPR use ISO just as manufacturer set it.. they are not testing "real" ISO accuracy.. but I can be wrong ofc..
Only problem I see with difference between real ISO and those set by manufacturers is, that you won´t get same exposure setting with different cameras..
In theory, if you use Panasonic G1 and Nikon D5000 and some nominal ISO camera and shoot same pic with same shutter speed and aperature value, you´ll get image exposed as intented on nominal ISO camera, but it will be slighlty overexposed on G1 (cause for example ISO 200 is real ISO 266) and slightly underexposed on D5000 (cause ISO 200 is real ISO 140). At least according to one-quite-trustfull site on net.. I never tried that.. so I can´t tell if it works or not.. biggest difference would be probably on highest ISO.. (which is anyway unusable on both:) ).
Posted by: Mescalamba | 23 Sep 2009 19:20:05
Nice blog post.. I´ll be happy to read part two. :)
Posted by: Mescalamba | 23 Sep 2009 19:22:16
I don't believe the illustration at the end is quite correct. Should not the RAW file be created prior to a/d conversion? It is actually the analog readouts of the sensor signals, no? Only formats such as jpeg and tiff happen after a/d conversion, that's why it's there.
Posted by: photomb | 23 Sep 2009 19:31:59
The diagram is misleading. When you change ISO you also change the exposure. Changing ISO frpm 100 to 200 make it possible to use half the exposure time or one step smaller aperture. Hence the sensor receives less light and the output from the sensor is lowered. Because of the lower output from the sensor the amplifier have to increase amplification to produce the same level of input to the A/D converter.
In the figure the signal from the sensor and amplification level should change, but the signal level from the amplifier to the A/D converter should remain the same.
Posted by: sula | 23 Sep 2009 20:12:05
photomb:
in digital electronics analog signals are always recorded digitally, signals need to be sampled and the precision level is the bit level the chip can provide; raw is the original digital record of the analog signal whereas jpeg is a compression format that is acceptable by many media, however any compression format will lose information. theoretically the better a processor-sensor partnership the better quality the raw file converted from.
Posted by: Tailai WAN | 23 Sep 2009 20:16:52
Mescalamba:
I said it backwards above. ACR show slightly less exposure than the camera.
I'm not talking about LCD exposure, but histogram, and even clipping highlights. With recovery off the image will show no clipping in ACR when it does show in the camera. And the Camera Histogram will show more exposure.
I long ago learned to ignore the difference and expose for ACR, but it's there.
Posted by: Kachadurian | 23 Sep 2009 21:53:00
Sula:
This is what happens if you take two consecutive pictures under the same lighting conditions at different ISO settings. In other words if you set the camera to "manual" and take one picture at ISO 200 and another at ISO 400 only the signal would be amplified. This in turn allows you to use a wider aperture or a faster shutter speed to achieve the same exposure.
Posted by: Chris767 | 23 Sep 2009 23:10:04
It seems to me that sensor dynamic range is at least as important as ISO for most picture-taking situations. Has ISO become like megapixels -- a marketing spec rather than a technical spec?
Posted by: Christopher | 24 Sep 2009 00:37:44
Kachadurian
Than I guess you are right, its different curve characteristic.. Tough. I don´t understand why.. cause ACR should use that within RAW.. instead of using its own.. Wierd..
Posted by: Mescalamba | 24 Sep 2009 00:49:01
Great post, it goes to show that you guys are, as always, paying attention and interested in what goes on in the forums. Great work!
Posted by: Sérgio Silva | 24 Sep 2009 01:28:04
Agree with Kachadurian.
Hopefully part II will bring more light to the issue.
BTW, I've noticed the exact same highlight clipping notice for the in-camera previews, which—if were marginal overexposure—once downloaded to Lightroom disappeared. I'm almost sure that this is due to the fact that in-camera previews are JPEG with limited dynamic range. Anyway, I always use the highlight clipping notice, still knowing its characteristics.
The other issue I was curious about, but couldn't find a definite answer to is this:
Is it better to underexpose and correct in software or use the higher ISO settings? And WHY?
Currently, I go the orthodox way—shoot with higher ISOs.
Lately, I've heard of someone who shoots only RAW at base-ISO200 on SONY A900 and created a software program (called RAW PHOTO PROCESSOR) to do the conversion at *home*. He claims he's achieving better results.
Is this possible?
Itchy to read the part two :)
Posted by: kendimian.wordpress.com | 24 Sep 2009 07:43:51
Wow. Great post. Thank you DPR
Posted by: bink | 24 Sep 2009 08:25:34
I've updated the diagram to show what happens if you just change the ISO (and not the shutter speed, aperture or light level). Hopefully this should make it clearer.
Posted by: Richard Butler | 24 Sep 2009 09:31:56
kendimian.wordpress.com
If you mean, that he converts his RAWs via his program and get better results than for example ACR (which isn´t hard, cause ACR isn´t good for A900) .. thats possible. Theres plenty of programs for RAW processing..
Underexposing is done mainly cause, if I´m correct, most sensors have dynamic range towards shadows, not highlights.. eg. shooting with slight underexposure should give you little bit more dynamic range, when corrected later..
Higher ISO lowers your dynamic range, but you get your pic exposed way you wanted.. personaly I shoot lowest ISO possible, cause it gives me highest dynamic and tonal range.. (and lowest noise ofc).
Theres few cameras that don´t need to worry much about ISO (one example - D700:) ). But standart APS-C usually need to worry about ISO..
Check your camera on dxomark.com .. or others.. you´ll see what I mean. :)
Posted by: Mescalamba | 24 Sep 2009 13:52:49
It is even more complex to find out how sensitive your camera is. Ther is another factor which comes to the equation, and that is the lens. You never know if F3,5 with one lens is the same as F3,5 with another one. And, as you go cross-vendor, you can´t use the same lens on a different brand camera, so you never know the truth, unless you measure the lens off-camera.
Once I have been shooting using my Nikon D70, side by side with a friend using his Canon 350D. I have noticed that he could go without flash, while my images were blury already. So we compared by using manual mode and identical apperture and exposure values, and the result was about one stop in 350D´s favor, as seen on the camera display and histograms. So what was it ? The sensor, the lens, the processing, the display, the histogram ? The only way you can tell is measuring things physically, maybe something like DXO does.
BTW, I believe that most concearned about sensor sensitivity are the astrophotographers, who count photons they can capture. You can learn a lot about your camera sensitivity and other hidden things by reading what those guys discuss.
For example, I have learned, that Nikon RAW on my D70 (and probably other Nikon cameras as well) is actually processed output for longer exposures, not really RAW. Nikon uses an algorithm to remove hot pixels, which in turn was removing stars from the astrophotographs. ;-) If you turn on the long-exposure noise-reduction, but switch the camera off while it computes the dark frame removal, you end up with a real RAW file (full of hot pixels, by the way) ;-)
Posted by: www.facebook.com/profile.php?id=100000101498979 | 24 Sep 2009 16:17:51
The best proof that the sensitivity of a sensor can not be measured is, that each camera only has one sensor, but adjustable sensitivity from ISO100 to ISO1600, for example.
Any sensor in the world can basicaly have any sensitivity you wish. But it comes at a cost.
The sensitivity must always be judged in relation to noise and dynamic range.
Native sensitivity of the sensor is given by many factors - IR filter transparency, antialiasing filter transparency, per pixel micro-lens light throughput, pixel size, sensor efectivity in terms of converting photones to electrones (how many photones gives how many electrones).
But you can always increase the sensor output by electrical amplification.
If you amplify the signal, you increase noise, so that´s the trade-off.
Posted by: www.facebook.com/profile.php?id=100000101498979 | 24 Sep 2009 16:31:29
Martin Grecner said:
"Any sensor in the world can basicaly have any sensitivity you wish."
That is not correct. Each sensor's sensitivity is fixed, and there is enormousl range between different sensors. Believe me, the Hubble Telescope's sensors are much more sensitive than any camera any of us could ever afford! The ISO scale is simply a standardized way of calibrating the amplification of the sensor's signal so that different cameras give the same output at the same ISO number. At high ISO numbers, sensors with low sensitivity require a lot of gain, resulting in intolerable noise levels. Sensors with high sensitivity require much less gain at the same ISO number, resulting in much lower noise.
Posted by: Christopher | 24 Sep 2009 17:17:08
kendimian.wordpress.com:
If you have to choose between seriously underexposing and using a higher ISO setting, choose the latter. Why? Because you'll get a higher S/N ratio.
Posted by: pzl | 24 Sep 2009 21:48:10
I think we should have a new type of reference for digital camera sensitivity. It works differently than film, to use the film ISO rating for digital camera seems not appropriate anymore.
I suggest every digital camera indicate it's base sensitivity, that indicates both the sensor intrinsic sensitivity and the noise it produces, measured using a fix (reference) amount of light, and a fix amplification.
For example, sensor A, exposed by 100 lux of light, amplify the electrical signal it produces by 100, and measure the result. Let say the result is 'X'. This 'X' would indicate the base sensitivity of the sensor.
Noise also measured, as well as the average factor the noise increases by increasing the amplification. For example by measurement above the average luminance and chromatic noise is 'Y', and it increases by the power of Z when the amplification is doubled, then the noise profile is Y^Z.
So every digital camera indicates it's sensitivity as : X/Y^Z
Let's call it XYZ profile of the camera :D
And the ISO dial is just simple indicating the amplification factor by the sensor's Amplifier, eg. +1, +2, +3, etc.
My2Cents :)
Posted by: Bram | 25 Sep 2009 06:51:16
Another thought, a better way to indicate the camera sensitivity maybe:
X / Y*A^Z
X : Sensor sensitivity
Y : Average luminance and chromatic noise level for standard amplification
A : Level of amplification
Z : Increasing noise factor relevant to amplification (maybe measured from amplification 1x to 16x).
And for the ISO dial, it should be change to linear amplification, such as gain level: 0 db - 40 db for example. Or x1, x2, x3 for step amplification.
Posted by: Bram | 25 Sep 2009 07:36:22
Christopher, you said that I was not true saying "any sensor can have any sesitivity", but you ignored my next sentence, which says that it comes at a cost.
And the amplification noise you are mentioning is part of the cost.
So we are both saying the same thing, I believe.
You are right saying that there are better and worse and terrible sensors regarding their capability of converting photons to electrons, and yes, this is what matters.
But unfortunately, this is what you can´t measure without taking the sensor out of the camera or without taking noise and dynamic range into equation, which I was trying to say by the following sentence:
"The sensitivity must always be judged in relation to noise and dynamic range".
So, I believe one must differentiate between ISO sensitivity (which is a relation between LV or ammount of light and EV - the resulting exposure) and native sensor sensitivity, which tells us how much electronic amplification/gain must be applied to the sensor output to achieve that ISO sensitivity. And it is clear that the higher the native sensor sensitivity, the lower the gain required, and the lower the noise.
Posted by: www.facebook.com/profile.php?id=100000101498979 | 25 Sep 2009 11:01:12
Bram said digital is different than film. This brings me to a fair example of what sensor sensitivity is and what amplification is, thinking in film terms.
If you think of film, the native sensor sensitivity can be compared to the sensitivity of the film emulsion. The amplification can be compared with for how long you leave the developing chemicals to develope the image, or what concentration of the chemical solution do you use.
If you leave the film develop for a longer time, it will result in a more dense exposure. But this off course increases grain and reduces dynamic range.
I believe there were commercially available films which had flexible ISO sensitivity achieved exactly this way. It was your choice whether you decide to expose it like ISO100 or ISO400, you just indicated this in the lab, and they have adjusted the development process to the ISO value you have choosen.
Posted by: www.facebook.com/profile.php?id=100000101498979 | 25 Sep 2009 11:17:18
Its a sad state of things when photographers try to be mathematicians.
In the end, it doesn't matter what numbers are used for whatever sensitivity - you're going to use whichever level suits your shooting. I mean, are you going to use inferior levels of ISO just to say to yourself that "lol, I'm sticking true to standards"? Do you buy cameras just to troll on forums saying such and such is inferiorly designed? Please.
Keep up the reviews the way they are, DPR.
Posted by: dfsaqwe | 25 Sep 2009 15:03:28
Hm...DP say, "we really can't measure true ISO, even though we used to do that...so we shouldn't, and you shouldn't care about true ISO either. Just trust the mfg. That's why we do camera reviews in the first place".
DxO say, "not only do we test true ISO and compare it to indicated ISO, but here's how we do it and here are our test results. If you want to see indicated ISO, SNR & DR vs true ISO, just look at the charts on our sites, which are freely available for anyone who cares to look, for every camera that we've tested. You can even compare up to 3 cameras at a time!".
Hm.
Which to believe?
I'll go with the data.
Posted by: Colossus | 28 Sep 2009 02:15:44
Just to make clear, this article (and the one that will follow, later this week), is a prelude to us re-introducing our tests of ISO.
In part I'm hoping that the article makes clear why we'll be testing Standard Output Specification - because that's the measure that the camera is designed to offer. I'm not suggesting that there's no merit in investigating aspects of sensor behaviour but it would be unreasonable for us to expect manufacturers to live up to a standard they don't use.
Posted by: Richard Butler | 28 Sep 2009 09:42:53
"I'm not suggesting that there's no merit in investigating aspects of sensor behaviour"
...but that's exactly what you do, when you take mfg ISO ratings for granted and base your reviews on them.
"... but it would be unreasonable for us to expect manufacturers to live up to a standard they don't use."
That all depends on the standard, doesn't it?
For example, it would be nice if all else being equal, a Canon 5D shot at say F8 ISO1600 would have an equal exposure to a shot at F5.6 ISO800. And so on.
Is that too much to ask of Canon?
And fair likewise, I think, of Nikon, Sony, Panasonic et al...
Now how about from model to model, from year to year?
And why should the results from a digital camera have no correlation at all to film results? Or does "film ISO" mean something completely different from "digital ISO"?
There's really not much room for error here. Sure, maybe a given manufacturer uses a different test methodology. But their results ought to at least be consistent across cameras within the same manufacturer and *also* show strong correlation to the results from other manufacturers. Otherwise there's no point in talking about ISO. It would just be a marketing term.
You ought to be able to take a shot from a fullframe DSLR with a given lens at a given Fstop and exposure time and compare the exposure directly to the results from a p&s, within reason. All you guys have been doing is giving a bunch of reasons to not do that. Instead of just doing it and explaining your choice of methodology. "The manufacturer doesn't do it that way" is just not good enough. Either they should do it your way, or you should change your methodology. A disagreement over technique is not a good enough reason to not do it at all.
Posted by: runrabbitrun | 29 Sep 2009 00:20:09
Great article. Let's not forget that the ISO setting is simply the gain on that amplifier shown in the bottom figure. Of course, the camera knows that setting and adjusts the exposure accordingly, which is not indicated clearly in the figure. It might be good to note in the last figure that the f-stop and shutter speed remain fixed in a manual mode setting - thus letting in the same amount of light produces a brighter image, as illustrated. In all but manual exposure modes on current DSLR's, changing the ISO changes the expsoure f-stop/shuttter. In that case, the wave coming OUT of the amplier looks the same, but the wave going IN does not. That is, increasing the ISO allows the output amplifier output to remain the same when the input has decreased. This is not clear in the figure as the image brightness would not change in any of the automatic modes (P, A, S, etc.)
Posted by: William S | 29 Sep 2009 16:27:57