« Photo Book Bestseller | Main | Supercool Case »

Wednesday, 26 September 2012

Comments

You write, "Almost always, they choose a camera ISO that is somewhat less than the sensor ISO because blown-out highlights are unpopular." This seems like backwards reasoning. A lower ISO means that more exposure is needed, and that means a greater chance of clipped highlights. Or I'm missing something.

Bill

Wouldn't the makers generally choose an indicated camera ISO that's _higher_ than the sensor ISO? That way you're underexposing a little and should be less likely to clip the highlights.

Nobody wants to hear that in the exact, precise, digital world, something like the ISO rating for sensitivity is determined solely by "whatever looks good," but that's how it is.

It's even worse than that, though. The ISO control on a digital camera controls two separate things: sensitivity and metering. The Exposure Compensation control provides a way to adjust the metering "exposure index" a few stops either side of the sensitivity setting.

The ISO standard for in-camera light metering and auto-exposure, ISO 2721:1982, hasn't changed in thirty years. It's still applicable to modern DSLRs. But... in the case of multi-zone metering systems like matrix metering and evaluative metering, the standard essentially becomes "whatever looks good."

So there we are: the metering and auto-exposure may be based on whatever the manufacturer thinks looks good, the sensitivity is based on whatever the manufacturer thinks looks good, and furthermore the in-camera JPEG conversion is based on whatever the manufacturer thinks looks good (trading off shadow noise for highlight headroom).

We can take some comfort that manufacturers can't be totally arbitrary and start applying just any old ISO numbers. The cameras are still expected to make exposures "that look good" when used in manual mode with settings obtained from a handheld light meter at the same ISO setting. (Handheld light meters are subject to ISO 2720:1974.)

Thanks for an interesting and informative visit to this topic, Ctein. ISO has actually been on my mind for a while now, the essence of my thinking being essentially that, as you conclude, the ISO ratings of a camera are basically relative to that camera. Using everything from a Canon S95 to a Phase One IQ160 I've certainly observed this phenomenon.

Which has recently (past 2 years) prompted me to often just let the camera manage its darn "ISO" setting, particularly the small cameras (ex: S95, RX100, OMD, X10, GRD IV) which seem to be able to dial-in more precise ISO settings than are available through direct adjustment (ex: ISO 154, 230,...) It takes some controlled initial experimentation to become familiar with a camera's overall exposure program behavior. But once I get the hang of it I generally set ISO to "Auto" for casual or grab shooting.

Yes there are plenty of exceptions, particularly when I want very specific behavior from a camera. But the final benchmark with me is the image result. That's all that matters. Digital cameras today can generally make better technical decisions on-the-fly than I can. That's part of what I pay for. Letting the camera auto-adjust the sensitivity of its sensor does not make me a sissy.

Three Questions:

Given that there's no basic physical standard to hold camera makers to account, how can camera buyers make an informed decision (with respect to camera ISO) when choosing from among the glut of offerings by different (or even the same) camera makers?

Are DxO sensor/camera bench-marks as reliable as the lensmakers' MTF charts?

Can "noisiness" (or its dearth) be made as a proxy for camera ISO?

Many thanks.

Reminds me that back in the film days you could (and people used to) adjust your meters to a speed different from the nominal film speed to account for individual preferences just as you explained. It is interesting that this possibility is now lost in the digital age. Sure, you have exposure compensation but it would still be nice to have the 'neutral' position of this being user selectable.

Ctein, please write a book entitled:

"Everything you think you know about digital cameras is probably wrong".

I could then spend many happy afternoons figuratively banging people over the head with it on gear forums.

Do include a list of non-sensicle phrases and mistaken adages. "Zooming with your feet" would be top of the list (I guess it would be possible for chimpanzees but rather difficult for humans) and "DOF control" (which sounds like a checkpoint for hat wearers).

You have neatly covered the zoom vs. prime issue and set the record straight on ETTR, and more recently you have debunked ISO and corrected the definition of dynamic range.

Can we some time get round to the misuse of focal length, DOF control, F stop and perspective ;)

Dear Bill and David,

Oops! You're right. Word salad-- I reorganized the sentence from first to final draft and forgot to change "less" to "higher."

Mike'll fix it shortly.

Thanks!

pax / Ctein

Hi Ctein, thank you for you article on ISO.
Base on your info, it seems that sensor ISO is the objective measure while camera ISO is not. So, there is no universal exposure setting even under the same lighting condition for 2 different cameras, am I right?
I've also heard that medium format digital backs produce noisy images even @ ISO 800, why is it so?
Thanks a lot!



Shouldn't the title be Why ISO isn't EI ?

I think I know what you mean, and I think I agree with you.

Bigger pixels (or more silver in the case of film*) = greater capacity for photoelectrons above the noise floor = less noisy image = lower ISO

But I think that if you used EI when you are talking about the representation of a level of light in the image or file and ISO when you are talking about the sensitivity of the camera it would be vastly less confusing.

I also think there would be a lot less confusion if the camera manufactures would just say "this camera has an ISO of X but you can set the EI to this range"

I recall Kodak and Ilford marketed films where the stated ISO and EI were different.

The advertisements for cameras with the headline "NOW WITH LOWER ISO" shall probably appear alongside news items about the devil's new ice skates.

*ask me why I think thin emulsion T-grain films suck

I might be confused. But thinking about DXO data, is it advantageous to have the sensor iso much less than the camera iso? So that the camera tells you its ISO 200, but it will only saturate like ISO 100. So you will set a faster shutter speed to avoid clipped highlights and hope that the Dyanmic range of the chip is good and see all the shadow detail to offer a good exposure range. If the Dyanmic range of the chip is bad than you might loose the black and the perceived exposure range would be limited.
Now thinking OMD-EM5 you have great sony chip Dynamic range, and your not clipping the highlights as your under exposing. So why would this be perceived as bad on DXO? I mean shouldn't a chip score higher if its Sensor ISO is less than the camera ISO? Or is it too depended on chip dynamic range and with the new sony chips where you can pull almost 5 stops out of shadows with little noise does it become advantageous?

I am just trying to keep it all straight. Thank you for your time.

I've been wondering about how to match the exposure of my digital camera to my light meter.

In the old days I would have done the
""shadow speed" is .... at an exposure that produces a density 0.1 density unit (d.u.) above film base plus fog"

So how do I figure film base plus fog for digital ??

I second Steve Jacob's comment: please write a book! I'd live reading it, and it would probably be a gear idea for you. You already know most of what you'd need to write. You've already written a lot of it here. For you, it would practically be free money!

A few comments of specifics in the article (at the risk of making nitpicky "power dB" type comments). I know you have a small number of words per article!

Ctein wrote: Suppose, though, you just make larger pixels that can hold more photoelectrons.

There is an ambiguity in the word "larger". It doesn't matter if you make the pixels larger if their capacity (literally) to hold photoelectrons scales with the pixel area as it usually does (more area; more photodiode; more floating diffusion (usually the limiting factor in CMOS sensors); more capacity). With a fixed number of photons per unit area hitting the sensor to get saturation so it doesn't matter whether the sensor is a single large photodiode or a 12 million smaller photodiodes if the capacity scales as the area of the pixel they will saturate at the same fluence.

Deleting "larger" makes it more clear that if you change the pixel design to make the pixel hold more photoelectrons, for example by making the photodiode deeper or the floating diffusion larger then, this is true: the saturation capacity goes up and the "Sensor ISO" drops.

Ctein wrote: Their pixels aren't insensitive, but they are designed to be huge light buckets, so it takes a lot of light to fill them.

It's true that medium format sensors (and up) have large pixels most have poor quantum efficiencies as they are older CCDs. Quantum efficiency is the probability a photon striking the sensor will create a photoelectron. QE is affected by PD depth; amount of obstructions overlying the pixel; and microlenses on the pixel. The old CCD designs really are less sensitive with QEs in the low 20% range compared to modern CMOS sensors that have have QEs approaching 60%. Even more interesting the photoelectron saturation capacity of these CCD pixels per unit area isn't good compared to modern CMOS sensors. This appears more clearly in the http://sensorgen.info renderings of the DxOmark results into sensor parameters where they derive the QE from various sensors. Compare the Pentax 645D and the Pentax K5. The small pixels on the K5 have much higher saturation than the huge 645D pixels. People who use medium format cameras don't use them for the sensor quality.

The QE is perhaps closest to what most people think of as the "sensitivity" of the sensor. With the latest Sony sensors it's probably hitting it's upper limit of what is possible with perhaps only 1/2 EV left to gain. We are starting to hit a plateau in that sensor parameter. We're almost there in the read noise. Don't expect future sensors to change as fast.

Ctein wrote: The noise in the picture is the result of the entire electronic chain, the amplifiers and all the rest.

This glosses over the three noise regimes for sensors: read noise limited; shot noise (photon count noise) limited and pattern noise limited.

The first occurs at low photon counts so at higher ISO and in the blacks. As Ctein showed last week the second is most important in the shadows, midtones and up are all. These are dominated by photon count noise not sensor or ADC or preamp noise. This is even more the case at base ISO on modern low noise sensors where read noise is tiny fraction of the photon count shot noise. This is an important point and one that is often glossed over in non-technical articles -- the midtone noise is set by the number of photons counted (and nothing else). That's why bigger sensors (of the same technical generation) win.

But the point is well taken the camera designer (even using the same sensor e.g. Sony sensor in the K5 versus D7000) gets to trade off shadow noise for highlight headroom. Where they place the midtone has some flexibility but as the saturation capacity contributes to read noise (so called kTC noise) the sensor designer puts some constraints on how far they can move that midtone.

Perhaps there's a future article on the three different noise regimes of imaging sensors waiting to be written for TOP.

For those really interested in understanding imaging sensors read Albert Theuwissen's blog articles at Harvest Imaging. He wrote the book on the subject (literally Solid-State Imaging with Charge-Coupled Devices in 1995).

Dear Sarge,

Answers:

1) All the camera manufacturers are choosing camera ISOs that produce the best overall exposure results for their camera. The only way you can decide if one cameras ISO 12,000 is superior to another camera's ISO 12,000 is to look at photographs from the two cameras.

While sensor ISO will have some relationship to image quality in low light, it's a very loose one. It is not a good criterion for deciding to choose one camera over another, especially if you're trying to nitpick the measurements to finer than one stop.

2) Are apples as reliable as beefsteaks? (In other words, this is a meaningless question.)

3) Not really. ISO is about what gives the best picture in terms of overall image quality. Noise is only one of many components of image quality.

~~~~~

Dear Chris,

It is in some cameras. For instance, in the Olympus OMD, under Utilities menu, there is a setting called Exposure Shift. That lets you dial in a permanent exposure adjustment for each of the metering modes.

~~~~~~

Dear Steve J.,

If you want to e-mail me privately with your rant, I'll be happy to read what have noted as the misuse of those words. I won't promise any substantial reply by e-mail but it might spark future columns. Please don't post it as a public comment; I won't respond and it would be way off topic.


pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

Dear Ash,

Objectivity Is Irrelevant. That's a fundamental philosophical misunderstanding that photographers have about ISO. An “objective” ISO is in no way better than a “objective” one. The sole purpose of ISO, always and for all time, has been to allow calculating an exposure that would produce a good photograph. Most methods for creating that ISO ultimately relied upon the judgment of folks looking at a bunch of photographs and deciding that some particular standard would, on average, produce the best results most frequently. The wording of the standard may appear objective, but that's not the underlying reality.

I've only used one medium format back, a rather ancient Phase One, but its performance at ISO 800 was extremely good. If you look at the sensor ISOs for medium format backs, they are often quite respectable. It's just they are nowhere as high as you would expect for something with 10 times the light gathering area of a lesser camera. That's because so much of that extra light gathering is going into filling up the pixel wells.

~~~~~~

Dear Hugh,

About ISO vs. EI, no, not really (Although I could've titled the column “Why ISO Isn't ISO Isn't EI.”) These are both ISOs. In fact, unless you are determining the exposure by the “spot meter for the diffuse highlight” method, camera ISO will give you the optimum exposure far more often than sensor ISO will. And that, after all, is the purpose of ISO.

Exposure Index is how you choose to deviate from the industry-standard to achieve the photographic results you personally want. That's not just you as the individual, but you as the camera maker. I do recall (unless my memory has deluded me) that in some cameras there are extremely high settings that are labeled as EI, because really they are nothing more than "push-processing," they don't conform to any ISO standard.

Your film memory is correct. Kodak TMAX P3200 had an ISO of 1000, but it legitimately had shadow speeds up to 3200. Why weren't those called ISOs? Because you only got that extra shadow speed with extended processing: P3200 emulsion contained low-speed silver halide crystals that developed quickly and larger high-speed ones that develop slowly. Normal/standard development gave you the official ISO with finer grain. If you developed the film longer, the high-speed crystals started to develop and the film speed genuinely increased (along with the grain). But those were not properly ISO speeds, because they did not use the standard development.

Similarly, Fuji once published a technical paper explaining why the ISO of Velvia was different from most photographers' preferred EIs, which had to do with the peculiar contrast and saturation characteristics of that film. Exposing it to a grayscale (the standard way of measuring ISO) produced a visibly different apparent speed than exposing it to saturated colors.


pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

Dear Kevin,

What you said.

I glossed over all that stuff very casually, because lay audiences are very easily confounded by excessive side details. And I was already at 1000 words.

The truth is that this ISO stuff isn't easy to understand. I only finally got my head wrapped around it early this month while working on the RAW columns, and it took quite a bit of head scratching by me, Oren Grad, and Mike before it sank in. If this column makes it seem to any of the readers that it is a simple and straightforward thing to understand, that's only because I worked hard at writing it that way.

~~~~~~

Dear Nick,

Thanks for the vote of confidence, but the narrative flow for that book still isn't there, and I'm currently working on another major writing project (no, I can't go into details). I haven't forgotten nor given up on the idea of writing the digital printing-photography book, but it has to come to fruition in its own time. I can't just order my muse to sit down and write it -- she takes direction terribly, and I am not the boss of her.

~~~~~~

Dear David,

I wrote the original sentence backwards: camera ISO is higher than the sensor ISO in order to produce less exposure, which gives more room for highlights.

Trade-offs between exposure range (NOT dynamic range, see my column of last week), highlight clipping, and noise are among the factors that enter into how good a photograph looks, so, yes, they affect what ISO the manufacturer decides on.

You've essentially got it right about the OMD, which means you are understanding this stuff. I'll be getting to the specifics in my RAW column, but the short version is that the additional exposure range in the OMD over the Olympus Pen is mostly allocated to the highlights. From black through the upper midtones, the RGB values in converted RAW files for both cameras track almost identically. There's a pronounced rolloff in the OMD highlight contrast which extends the range there and gives a much more “film” look to tonal rendition.

(Folks, please don't ask for more details now. Wait for two weeks for the full column.)

~~~~~~

Dear Robert,

Well, how do you do your metering? It's the same question you would've had to ask in the film days, if you wanted to match your film camera's exposure meter to your handheld one. There's no ab initio, objective way to do this. You do it by running tests and see what matches up.

For instance, if you were the kind of photographer who metered off of the shadows with a spot meter, as a few did, then your personal EIs for films were probably a factor of 10 greater than the official ISOs, and you adjusted your meter accordingly. Contrariwise, if you spot-metered off of diffuse highlights, as an uber-serious transparency photographer might, you substantially downgraded your ISOs to get your EIs. And, if you were doing averaging metering, you might've tweaked your personal EI up or down by half a stop or so to bring your equipment into synch, but not much more.


pax \ Ctein
[ Please excuse any word-salad. MacSpeech in training! ]
======================================
-- Ctein's Online Gallery http://ctein.com 
-- Digital Restorations http://photo-repair.com 
======================================

May I third the request for a book on digital sensors and photography. As Nick said above, I think a lot of the guts of it is already written here in many of your columns though that doesn't mean that I think it wouldn't involve some more work.

So, Mike could consider starting the TOP Press and taking pre-orders to fund your time while completing the book. I, for one, would happily be prepared to contribute now for the promise of a book in the future, tomorrow preferably but next year would be fine.

(Just joking about tomorrow. I remember how long it once took me to co-write a paper for a technical journal and re-edit following the peer panel reviewer comments. Writing is not a quick or easy task and I appreciate the time and effort which go into columns like yours).

Kodak really did double their film speeds when Tri-X went from ASA 200 to ASA 400, the claimed removal of the safety factor was a ruse to throw off the competition. See: http://www.apug.org/forums/forum205/39398-how-did-speed-jump-old-films.html

Of course, DuPont, Ansco, Ilford, Agfa, and all the rest surely analyzed Kodak's new ASA 400 Tri-X, and realized there was a real difference.

If you are really bored you can search on the Oly dpr forum for threads concerning the E30 and E620, which had about one stop sensor/camera difference starting at iso 200. Cameras before these had close sensor/ camera iso's. The threads go on forever. Some folks were angry at the degradation of shadow detail. DPR was very impressed with the improved highlight control, especially on the 620. Anyway, the argument has died away mostly thanks to the much improved OMD sensor (except that it still seems to enrage a few folks who think Oly is somehow lying to consumers).

Dear John,

If Ron Mowry says it, it must be true.

Seriously. This is the guy I go to when I have technical film questions. Well, him and Dick Dickerson and Sylvia Zawadski. Between the three of them, they know everything there is to be known about emulsions.

Thanks for the corrected information, much appreciated!

pax / Ctein

Of course ISO isn't ISO. It's ASA!

I own the GH1+20mm and taking pictures of my family indoors I prefer to use 1/80s which sometimes means ISO 3200. Naturally I'd like to know if the E-M5 would give me better image quality under the same conditions (same lens, same shutter speed).

It seems both cameras would require different ISO settings to achieve the same exposure, so comparing online samples at the same ISO is somewhat meaningless. Which I think is the reason people get upset about it.

So, If I want to know if the E-M5 has, lets say "better low light performance" than the GH1, is it possible to tell without trying the camera first? What numbers/ratings should I look at?

(I hope that makes sense, photography is not my main hobby)

Dear Mart,

If two different cameras are doing their metering the same way (a big if) and if they have both been properly calibrated, they will give the same exposures to the same scene at the same camera-set ISO.

Give or take half a stop, in reality.

Equipment just isn't all that accurate.

Metering's a big if. Multi-segment smart metering varies the exposure compensation from scene to scene, to try to give the most satisfactory results. Every company has its own proprietary schemes for doing this. So, Camera A might look at a backlit scene and open up by 0.5 stop and Camera B might look at the same scene and open up by a full stop. But that's not an ISO problem.

So, yeah, there's no reason you shouldn't be able to compare online photos made at the same ISO, as much as you can ever compare photos.

pax / Ctein

A huge thanks to John Shriver for is link. This is a major breakthrough for me at least.

It seems that the key notion that emerges here is the SAFETY FACTOR. This is a concrete concept that should be made operative.


This project will allow us to better control of our gear in the field (Sur le motif magnificently said Paul Cezanne, a forgotten photographer from the pre-digital era), which is much more important than changing hardware every 6 months.

Ctein Be our AA, our Fred Picker, our Phil Davis and give us the BTZS of the digital era.

Dear Ash,

Oops, sentence should have read, "An “objective” ISO is in no way better than a “SUBJECTIVE” one.

My bad.

pax / Ctein

Mart, it is possible to work it out from the combination of DXOmark data and DPreview information.

One thing that dpreview does test is what they call ISO accuracy. In fact what they are measuring is how the metering system is calibrated. Fortunately, this lets us see how similarly two cameras, with the same lens (and therefore light transmission) will meter the same scene.

In your case, they report that the EM-5 meters to ~1/3 stop darker than a light meter would indicate, and the GH1 is probably similar (couldn't find it).

What I'd then do is take the DXOmark data for screen level 18%SNR for each iso, and compare them, not as DXOmark does at their tested ISO, but at the manufacture stated iso (-1/3 stop for the EM-5) to see which gives a higher SNR. You can do the same for dynamic range. This should tell you which will be less noisy given a particular scene and lens.

Oh how I wish dxomark gave you the option to plot their results vs stated ISO rather than their measured ISO. Would be soooo much better for camera comparison!

So, If I want to know if the E-M5 has, lets say "better low light performance" than the GH1, is it possible to tell without trying the camera first?

Mart: it is better to accept that any statement containing the word "better" has no definitive meaning (grin). Sorry, but let's get that out of the way.

The same applies for the "low light performance" part of the query. This describes a very different mix of qualities, for different people, depending on what assumptions they are making, and on how they are judging the outcomes.

You can get a good indication about this by reading several professionally done reviews, or user reviews, from end to end, with your own requirements and circumstances borne always in mind. That's all.

Because no single technical number will be able to express such an ill-defined thing; to take an analogy, no single measurement can sensibly express "human physical performance". How applicable could that be to any given case? Say, if person A has the ability to run well, but cannot see. And person B has lots of muscle strength but no stamina. We'd need to know all about those things each time, in order to evaluate just how misleading the standard test was each time.

Even if there were consensus that some technical test of photography in low light conditions was conclusive (hollow laugh), that is still just one item of disconnected evidence.

Wow, I did not know this.

I think this may explain in part why I am so pleased with the performance of my small sensor cameras (FZ150, G12), when DXOMark gives them comparatively low ratings.

As a fulltime freelance writer who often shoots photos to illustrate his stories, I tried to research and absorb a lot of the theory of digital cameras before making buying decisions. Your columns have been very useful and illuminating.

So I will underscore what others have said: you should write a book. To avoid the high cost of getting from silicon to paper, you could self-publish as PDF files on a CD ROM and sell direct. That would keep the cost of goods low, and you could publish on an "as needed" basis.

(That's what I did with my book "Elliott on Airguns" and the strategy worked -- I did not have to go into the hole thousands of dollars in the hope that the resultant pile of printed books would eventually sell.)

Of course, if a publisher wants to give your a juicy advance . . . all the better!

Still, I think your book would sell, even if the market might be small. But if you keep the production costs low, it could be a successful -- and well received -- venture.

The comments to this entry are closed.