A Proposal To Redefine Digital ISO

By Ctein

The definition of ISO is based only partly in the physical sensitivity of film or sensor. ISO is a judgment about what exposure will produce the best results, on average, with a particular medium of capture. ISO is a compromise.

In 1955 Tri-X had an ASA (ISO's predecessor) of 200. In 1965 it was 400. The film hadn't changed, just the official judgement of what the best exposure was. Higher exposure of B&W film produces better shadow detail, but increased grain and poorer sharpness. Shrinking film formats increased the importance of grain and sharpness in overall quality. The manufacturers decided the emphasis was too much on shadow detail at the expense of those, so ASA values were redefined. Negatives need sufficient exposure to produce shadow detail while not overly compromising grain and sharpness. Slide films must record sufficient highlight detail without having a slide that's unacceptably dark. Many negative photographers routinely "overexpose" by around a stop*. Many slide photographers "underexpose" their photos. We're all aiming for adequate information at the end of the luminance scale where the medium is most likely to fail us. ISO doesn't require you make a certain exposure; it's merely a best guess. But when ISO consistently guesses poorly for a majority of users, it ain't doing what it's supposed to.

Digital sensors have problems similar to slide films; expose more than a certain amount and there's simply no detail. Unlike slide film, digital sensor's behavior in the highlights is not at all graceful. Once the pixel's potential well is full of photoelectrons, clipping occurs. In a purely linear device, that can happen 2–3 stops above middle gray (there are ways to improve that—many cameras do much better).

At the same time as the highlights are getting clipped, there is usually detail well into the shadows, albeit with increasing grain/noise. Cheapo digital cameras can capture a seven stop range; better ones 9–11 stops. In many cameras the luminance capture range is weighted too far in favor of shadows. Especially true as we notice loss of tonal separation in highlights a lot more readily than we notice it in shadows.

From the comments I read about digital cameras (and my own modest experience) I think the digital ISO is simply too low in many (not all) cameras, often by more than a full stop. Many of us have already learned to "underexpose" as normal practice to more usefully balance highlight and shadow detail. We just set our exposure compensation dial for –1 or –1.5 stops and leave it there most of the time. This needs to be institutionalized. Digital ISO is due for reformation to provide exposures that more usefully exploit the sensor's exposure range, balancing adequate shadow and highlight range. On my Fuji Finepix s6500, changing the current ISO of 100 to ISO 250 would be about right.

Both of these photographs were originally made in RAW mode by my Fuji FinePix s65000 camera. (Click on the images to see them larger.) The one on the left was made at ISO 100 with no exposure compensation. The one on the right was made with –1 stop exposure compensation. Both of these photographs were originally made in RAW mode by my Fuji FinePix s65000 camera.

The one on the left was made at ISO 100 with no exposure compensation.  The one on the right was made with –1 stop exposure compensation. I made a curves adjustment in Photoshop to the image on the right, so that the shadows would have about the same brightness in both photographs, for comparison purposes.

The "normally" exposed photograph has blown-out highlights with little or no data in one or more channels. The "underexposed" photograph has good highlight detail and only a slight loss of clarity and separation in the shadows. So, which one would you say was correctly exposed?

There would be side effects. A good one is that you get more low-light capability. Instead of hand-holding at 1/10th of a second, you're hand-holding at 1/25th of the second. The bad is that your pictures become noisier. It's a balancing act, just as defining B&W negative ISO was a balancing act between shadow detail, grain, and sharpness.

But, digital image noise is already better than comparable speed 35 mm film in many digital cameras, and it continues to improve with each generation of sensor. Luminance capture is definitely not better than film in the highlights. I understand the current obsession among the digital newbies for low noise; I'm a fine-grain lover myself. But poor tonal placement is a worse problem that will improve less as sensors get better; it's more inherent in how the sensors map luminance values. We need a new ISO definition and we need it now.

_____________________

Ctein

*(By the way, this has nothing to do with the misconception that manufacturers overrate negative films by half a stop or so.  That's mostly due to mistakenly metering off of 18% gray cards using ISO-standard meters calibrated for 12% reflectance.)

Comments

[Chris Norris]

The "shoot underexposed" mentality depends a lot on how you're doing your metering as well. I find that for a lot of things I shoot, shooting with a +1/3 or +2/3 EV is correct. For my camera manufacturer to redefine 0 a little lower than it is now would force me up into the 1 1/3 range... pretty high if I'm maxing out at 2.

Basically I'm still waiting for digital sensors to do something special. Your last paragraph sums that up: things keep getting better with each generation of sensors. I can't wait for the sensors that come of Kodak's new Color Filter Array for increased light sensitivity, sensors that have amazing dynamic ranges... these things I will wait for. It's also why I'm baffled when someone with say, a Canon XTi will fret about whether they should upgrade to the 40D. There's not enough of a difference! Hold out as long as you can!

[Till]

An experienced member of my camera club suggested exactly this permanent -1 exposure compensation to one of the less technicaly inclined members to help avoid blown-out highlights. My own approach to exposure compensation has so far been guided by Mr. Reichmanns suggestion to expose to the right (http://www.luminous-landscape.com/tutorials/expose-right.shtml). Checking the histogram should prevent blown-out highlights. I agree that keeping shadow detail may be overrated when compared to highlight clipping, but don't you think that a constant -1 exp.comp. would unnecessarily lead to higher noise levels?

Cheers, Till

[Ken La Rose]

Dear Ctein,

I'm glad to see a discussion about "proper exposure" with digital cameras.

For me to comment on the “proper exposure” of a digital image I would like to see the histogram of the RAW image. Without the histogram information there is no way to decide on the “proper exposure”.

Isn't it easy enough to check the digital camera meter by exposing an evenly lighted surface at EV 0 (no exposure compensation) and looking at the value in PhotoShop? If the RGB values is are in the range 110,110,110 to 130,130,130 then meter is performing adequately. If the value is not within range then a series of +/- exposures can determine the EV value that will bring the meter into range.

Thank you for your discussions. I enjoy them very much.

Ken La Rose

[Doug Pardee]

We already *have* a new ISO definition, one which gives manufacturers free rein to specify ISO however they want as long as they label it REI (Recommended Exposure Index). Canon, for one, has already stated that all of their DSLR ISOs will be stated as REI.

Recent Canon DSLR models (1DmkIII and 40D) provide an optional feature called "Highlight Tone Priority" which reduces the "middle gray" sensor output level by about a stop. One could wish for the ability to specifically set where the middle gray level is, but the HTP option is a good start.

[cyril]

If I understand this correctly, this seems to go against the usual advice to "expose to the right", which tends to overexpose (as long as no highlights are clipped).

One usual complaint against DSLRs used to be that they mostly seemed to underexpose, in order to preserve highlights. Isn't that what you suggest?

In fact digital ISO as implemented for example on the K10D is essentially another axis for controlling exposure : do the actual numbers really matter that much?

[Tom Darnall]

Okay, now I'm really confused.

About a year ago, I read an article on Michael Reichmann's web site called "Expose To The Right," and have incorporated his suggestions into my practice.

It says you should "bias your exposures so that the histogram is snugged up to the right, but not to the point that the highlights are blown." For many scenes, this means the image will look too light in the RAW converter at the outset. So, it requires pulling down the exposure when doing the RAW conversion to get the look you want.

Supposedly, this helps produce the best possible tonality in the photo, and the practice is based on the fact that there is a more information captured in the upper regions of the histogram by a long shot. The capacity to record information, working right to left across the histogram, is as follows:

- Within the first f/stop, which contains the brightest values, there are 2048 levels available.

- Within the second f/stop, there are 1024 levels available.

- Within the third f/stop, where the middle tones reside, there are 512 levels available.

- Within the fourth f/stop, there are only 256 levels.

- Within the fifth f/stop, where the very darkest values are, there are only 128 levels.

I know you have been a guest on Michael's videos before, and I'm wondering if you guys have discussed this. I'd be interested in how these two ideas for exposure are reconciled. Perhaps one argument is the other side of the same coin...don't know.

[Hugo]

"Balancing adequate shadow and highlight range" - as you say. This is the key for getting the most out of a digicam like the S6500fd. I use the same camera for over a year now. Recently, reading a review about the Sony A700 (http://www.akam.no/test/speilrefleks/akam_tester_sony_alpha_a700/43884/5), I remembered that the S6500fd is shipped with iso 200 as default iso-setting. The Sony A700 seems to have the best dynamic range at iso 200, and maybe it's the same with the S6500fd. Now I use to shoot at iso 200. Since I use -0,67 exposure compensation too (in order to avoid blown out highlight), I gain roughly 1.5 stops in speed. Results with iso 200 seem more pleasurable, presumably with a slightly better dynamic range, though I'm not able to measure it. Noise wise I see no problem shooting at iso 200 with the fuji S6500fd.

[Michael Tallman]

An interesting discussion. ISO to me seems like kind of a loose standard to begin with. When I would shoot film, for instance, I would expose a 100 iso neg film much differently than a 100 iso slide film, based on the particular qualities of those films. Digital SLRs have their own particular qualities of how they use ISO and how it relates to exposure - taking the time to really get to know your "materials" is worthwhile. Do we need a standard for this thing that has only been loosely followed for so long?

I use a lot of aperture priority on my 5D watching the histogram to optimize my exposures for maximum information. I can do more with my prints from files that are "exposed to the right" just like I could do more with my silver prints with a well exposed and developed negative. I've found that the "0" midpoint is in a useful spot as far as how often I need to go beyond -2 or +2. If I were to go about making a print from a raw file the two images you put up, I would try the image on the left first, as long as the sky is not blown out - it doesn't appear to be but I can't tell.

[Eduard de Kam]

The funny thing about digital cameras and ISO is, that when using a separate meter, the only way to get the 'right' exposure, I find most camera ISO values are to high. I read the light as it falls on the white dome of my Minolta meter. My Leica M8, at ISO 160, likes the meter to be set at ISO 100, or 80, depending on the situation and the light.

[Donovan Rieger]

I assume that sensor sensitivity is expressed in terms of ISO because most digital photographers originally shot film. But why not just refer to sensor sensitivity as low, medium, high, very high and ultra high? Of course, that would make it hard to use a hand held meter. Still it seems that with digital we need to get away from the film paradigm in all its forms, and accept that digital imaging is different in very many ways. It requires its own language without constant comparisons to film and film cameras.

[Tim]

I'm afraid this article pushes all my wrong buttons.

First, the talk of over/under-exposure is completely lacking any standardised reference-point. You need both incident and contrast measurements of your scene in order to have a meaningful discussion. You can't just point at a picture and start talking about under/over-exposure.

Second, there comes the old old issue of what correct exposure might be. It is simple. "Correct" means "where you wanted the tones placed" and is therefore completely subjective. The camera, with its various computation algorithms, does not define correctness, merely approximations to things you will likely find acceptable. I thought we would know this around here of all places, but this article does not reflect as much.

For what it's worth, I agree with the "align right" approach. 99% of the time I'm in M+spot mode because it allows me instant measurements of highlights if I want, contrast across the scene if I want, correlate with sunny-16 if I want, and flexibility to vary either shutter or aperture to place the tones where I want in the histogram myself (highlights on the brightest end before it overreads) - it doesn't mean I've found a midtone and spotted off it! The remaining 1% of the time comprises very short attempts to experiment using matrix metering across the whole scene - I've tried, I invariably lose a shot in the wild as it simply doesn't know how much to clip highlights, and instantly flip back to M+spot.

Compensation is meaningless. Not only does it not fit in my philosophical model, but it presupposes that the highlights in your scenes will be a fixed N stops away from the camera's computed midtones. Unless you shoot only studio shots of test-cards in fixed lighting, that will not be the case.

[dasmb]

Tom --
"Expose to the Right" means "Get as much data in your exposures as you can without blowing your highlights."

What CTein is saying is that many cameras naturally expose too far to the right, and at times it's necessary to dial to the left to avoid blowing highlights.

The two statements are not mutually exclusive -- the goal is to get your histogram to line up so that it's just touching the right side, because if you take the opposite extreme you'll get harsh tonality.

In fact, I would like to posit that ALL digital cameras should have a mode that exposes in this way...and then uses software to dial the tonality back to middle grey.

As for ISO, I don't think it matters what the numbers on the box say -- you're going to have to learn how your camera produces images. My XTi definitely likes to blow highlights in scenes where I want to capture the sky, no matter what metering mode it's in -- and I've learned to dial back 1/3 to a full stop to get just the right balance between a dark foreground and a pinkish blown highlight sky (ew).

[Bob Dales]

Tom raises some good points. Looks like an exercise is in order. I'm going to take some shots both ways and see what works for me. Ctein's solution seems better for street photograrhy where you don't have time to study histograms.

[Ken]

Ctein: I agree with the spirit and conceptual propulsion behind your proposal to redefine ISO with the new, maturing medium of digital imaging in plain view. (Side note: Not long ago I was shocked, shocked! to learn that the specifications for ISO are proprietary and that one must pay a not-too-small fee just to obtain a copy.)

Spirit aside, however, I would be concerned that such an initiative would feature a PGR* far greater than 1.0. ISO ratings, as they exist, are at least consistent. The vast majority of digital camera owners have no idea what an "ISO" rating is. (I think many would speculate that it refers to the desirability of a camera, as in "I SO want that new camera!" Maybe they'd be right, too.)

Regarding the "expose to the right" philosophy....
I believe that this mantra came from work done by Photoshop's daddy, Thomas Knoll, and the late Bruce Fraser. Far be it for me to dispute the foundation for this theory: that the majority of usable information in a digital image lives in top two stops of an image (whatever that means).

If "expose to the right" works for you, terrific. But it never worked for me. Today, when controlling the camera's exposure manually, I will typically dial down 1/3-2/3 stop from what would be a normal exposure. Of course there are many situations where I must also dial UP the exposure. But this "expose to the left" bias has been a very positive technique for me.

--------
*PGR: Pain to Gain Ratio. The ratio of the general discomfort of having/doing something over the general expected benefit of that something.

[Carl Leonardi]

ISO is sort of meaningless in the digital world--The more sensitive you want the photo cell to be the more noise you get. With film you needed a point of reference to put on a light meter. With digital you have no need for a light meter.
Everything is sort of automatic in reference to the photo cells inherent design.
It's basically how much noise you can tolerate in order to get the photo you want. Yup expose for the hi lights and or use photoshop for the final results. Digital isn't film, the whole thing is very simple, shoot pictures not the bull.
Putting Film speeds on a digital camera is a bridge from film to digital but really not necessary in the real world of digital photography.

[J Ollinger]

I'm new to digital so I find all this particularly interesting.

The way I understood this in Film, I thought we were supposed to do the homework and find out how a given film reacted to a given exposure and development combination; i.e. Zone System calibration.

Is there any kind of equivalent for this in digital?

[Alan Rew]

Very useful article. I'll have to do some experiments with my Canon 5D now.

One technical issue with the on-camera histogram is that it isn't (AFAIK) a histogram of the RAW data, even if you're shooting RAW: it's based on an in-camera jpeg produced from the RAW data, in the colour space that you've configured in your camera settings for jpegs (e.g. sRGB, Adobe RGB (1998)). So it's only a rough guide to whether you've got blown highlights.

There was some discussion on the now-defunct RobGalbraith forums in which several people requested the ability to have a 'real' RAW histogram, but Chuck Westfall, the Canon USA technical rep, said (I'm paraphrasing here because I've lost the original discussion link) that such a feature wouldn't be meaningful. He didn't explain why. Maybe you could ponder on this, Ctein?

If anybody knows of any camera that definitely gives a histogram of the RAW (i.e. monochrome, un-processed) data I'd be very interested.

Regards,

Alan

[blurrylens]

All of this discussion sounds a lot like what I learned way back when in Photography 101. "Expose for the highlights, develop for the shadows" or was it the other way around? You know how in Adobe camera raw, there's the red and blue gamut warnings?

Well, the big day in digital photography will be when each pixel on a sensor can set it's own exposure and even it's ISO sensitivity individually to avoid being out of gamut. As it is now, digital sensors merely mimic film in being evenly sensitive across the entire surface. This barrier needs to be overcome.

As a photographer who focuses mainly on landscapes, I personally spend far too much time in Photoshop just evening out the tonal range or combining bracketed exposures. It will be a great day when this is no longer a concern.

[Damon]

I'm with Tim in finding this article philosophically baffling. Automatic exposure metering has always been about guesswork by the camera with of course no knowledge of what the photographer actually wants (and whether the scene elements are supposed to appear bright or dark). Matrix metering attempts to add that bit of artificial intelligence, but like machine translation, too often makes the wrong assumptions.

Expose to the right is a fine idea, but why should photographers have to do all the work? I'd propose that camera manufacturers add an 'expose to the right' mode that would adjust the exposure specifically to be as far to the right without clipping, and not worry about middle grey. This should certainly be possible now that even SLRs feature live view so that histograms can be seen even before the shot. In effect, this would use the sensor itself as the meter and lead to more exposures that meet the needs of photographers - at least the ones who understand the concept of exposing to the right.

Anyway, regarding this article, I don't think ISO values need adjusting so much as the cameras' light meters themselves - and a quick look through photography forums will show that cameras' metering systems already behave differently from one another, depending on model - some tend to underexpose, and some the opposite. The more I shoot, the more I'm inclined to agree with Tim's M+spot method.

[Stephen Best]

The image on the left is "correctly" exposed. That is, if you held a transparency up and compared it to the original scene, they would match. The same for a digital capture or negative you print with default/starting settings. Whether the highlights are blown or the shadows plugged isn't pertinent in the determination of whether it's "correctly" exposed. The latter are products of a mismatch between the camera and subject ranges.

This however is different from an "optimal" exposure where you treat the film or data captured as preserving information in the tonal range that you require, usually highlight detail for positive processes and shadow detail for negatives. For this you need to meter in the area of interest (with a spot meter or just using the histogram on your digital camera) and redistribute the tones in post processing. Just dialing in a fixed compensation may get you better results for a certain camera and certain subject contrast range but it's rarely going to be "optimal".

There's nothing wrong with the definition of ISO though it's starting to become increasing irrelevant for cameras with auto ISO functions and little penalty for high ISO settings.

I seem to remember a guy called Adams or something wrote a book about all this way back when.

[Mike]

"With film you needed a point of reference to put on a light meter. With digital you have no need for a light meter."

Carl Leonardi,
Not so! With digital you've still got to match the gain setting you've chosen with the exposure settings that match it. If you try to use the aperture and shutter speed settings for ISO 800 when you've got the camera set on ISO 200, or vice versa, you'll get an unusable digital picture 99% of the time.

Mike J.

[Mike]

"...i.e. Zone System calibration. Is there any kind of equivalent for this in digital?"

J Ollinger,
Sure, there's Carl Weese's "Paper Towel Test." It's pretty simple, too. Try here:

http://theonlinephotographer.blogspot.com/2006/12/
on-testing-new-digital-camera-steps.html

(You'll have to put the URL back together again, sorry.)

Mike J.

[john robison]

Can't some of these doggone machines just go through 5 rapid shots, -1, -1/2, 0, +1/2, +1 EV, and then you pick the best? Can elements of several exposures be combined? For us computer dummies, well, we think computers can do almost anything.

[Ctein]

Dear Folks,

WAAAAY too many posts for me to respond individually; I hope you'll understand.

First, addressing some specifics, then a comment on a general misunderstanding of what ISO's all about. Don't start composing your rebuttal until you've read this whole post.

Specifics: yeah, my rule is similar to M.Reichmann's-- expose as much as you can without blowing the highlights. Results will depend upon specific camera. Some will be high, most will be low (as mine is), a few will be "just right".

Photo on left in my illo has two blown-out channels in the RAW data over most of the sky, and one in the highlights on the leaves. Photo on right has no blown-out highlights, save some pinpoint specular highlights. The photo on the left is highly suboptimal, by any sensible measure. Just so's ya knows.

General remarks-- many of the commenters don't understand what ISO is about. It has nothing to do with your personal metering techniques. Everyone who has posted about how they personally meter a scene to get the 'correct' exposure is, I have to say, missing the point.

(Tim-- please take note: I put "over-" and "under-" exposed in quotes for a reason. Grok the reason and you will grok all.)(

Stephen-- The photo on the left does NOT match the scene. I was there, you weren't; take my word for it. And even if it did, it's a crap exposure.)

ISO is not "loosely" defined nor applied-- it is merely a starting point. Fundamentally, it was derived like this: go out and photograph a whole buncha different scenes using an 'averaging' reflected-light meter, using a range of E.I. settings on each scene. Now look at the results-- pick the E.I. that give the greatest number of "good" photos. That's your ISO. This is what underlies the ISO film/sensor speed standard.

The wonderful thing is that this works most of the time; the majority of scenes, no matter what they are, have similar reflectances. Empirical miracle. WHY? I dunno! But it's true. Else auto-exposure would have crashed and burnt from the get-go.

Obviously "good" is something that has to take into account many factors. The position of my column is that I think "good" should maximize the useful tonal range and minimize highlight blowouts, primarily.

OK, back to to y'all.

pax / Ctein

[Jeff Henderson]

John, The short answers to your questions are Yes and Yes. Most if not all DSLRs and most 'Pro-sumer' digicams can exposure bracket at least three shots +,0,- and some 5 shots. These exposures can be merged together to create a High Dynamic Range or HDR images. Photoshop CS2 & 3 can perform this merge semi-automatically. Photomatix is a popular application for creating a HDR image from several exposures and then tone mapping them so they can be displayed on our low dynamic range devices such as computer monitors and ink jet printers. Uwe Steinmueller who runs The Outback Photo web site has a tone of information on his site about this topic. Here's a link http://handbook.outbackphoto.com/section_hdr_and_tonemapping/index.html

[Eric Jeschke]

Expose to the right works if the scene has such limited contrast range that it fits within the dynamic range (think histogram) of the sensor. But many scenes are outside that range, and just like in the "good ole days" the photographer (gasp) must make the determination of what subset of that range he wants, and how it should map to the sensor's range (remember shooting transparencies?). So then you can shift it to the right and let the highlights and shadows clip where they will.

Then there is the question of how much post processing you want to muck around with. I know most photo pundits would argue that it's all just part of a normal RAW workflow, but...some of us don't really want to spend a lot of time for each photograph in postprocessing. Shooting to the right forces extra time moving the tonal range back to where you thought it belonged in the first place.

[Jobo]

Doug Kerr has several interesting articles on metering. In SOS_REI, you can read that the ISO since 1996 allow camera manufacturers to decrease the highlight headroom by half a stop (as long as they call it "SOS" and not "speed"). With ISO REI all bets are off, as Doug Pardee mentioned.

John: They can indeed, look into what Uwe at Outback Photo calls high-speed HDR. A big issue is that you collect motion blur from the moment you start the bracket until you are done, which limits the usefulness of the technique even for landscape photography.

And the Pantax K10 can even combine RAW exposures in-camera...

[Martin Doonan]

Interesting discussion, interesting article.
My take (FWIW): usually have to expos my Canon 20D +1 stop, because I main shoot bright scenes. With a scene such as Ctein shows, I'd be looking at -1 stop (thereabouts) due to the high proportion of dark subject. In mixed lighting, I just set AEB +/-0.5 with a -0.5 compensation and shoot 3 of everything.
The most important thing I found about ETTR (when shooting RAW) is getting the JPEG settings correct to ensure the histogram shows what you'd expect (see my tests http://doonster.blogspot.com/2007/05/great-eos-metering-test.html).

I think meters are still doing what they always did, trying to put the average of the scene luminance at the centre of the capture medium response. As some have said, a proper highlight priority metering would be better. With a matrix meter this would mean simply putting the maximum metered zone to the right, say +3 over mid: everyone happy.

[David Bennett]

Dear Ctein
I have a question for which I have a practical answer because I tested it, but for which I do not have a theoretical or technical understanding. With film I would know the answer, but I don't know what is going on with digital sensors in terms of increasing sensitivity. The question is this.

Using a Digital SLR
If I shoot at 400ISO, is this the same as shooting at 800ISO and putting in +1 exposure compensation?

[fernando]

I suppose this is all a bit confusing, and I wonder if it stems from digital trying to mimic analog measurements/sensitivities. To wit: digital capture should be constructed from the ground up. Meaning, since the camera can associate the measurement of light to the noise in a pixel, just adjust simultaneously the aperture/shutter-speed for the best signal-to-noise ratio.

Like many instruments, isn't it all about the best SNR? Analog cameras could do this, and more so if the film's ISO is set to some criteria that is not related to SNR (as noted in the article). However, whether with spot metering or evaluative (weighted) metering, it is possible for a digital camera to look at the darker and brightest area and determine an ISO/Shutter-speed if shooting with Aperture Priority -- I believe this is already happening.

Of course, so removing ISO from digital cameras now means that we have to dial in a "Grain factor" -- like when shooting manual. Oh dang, that is the way that I think of the ISO these days with digital -- if I want noise, dial up the ISO and if lower, then 100. It is only a relative measure and not absolute standard. So, they can change the ISO to a noise-tolerance dial.

The crux is going to be the algorithms to determine the best noise-tolerance given all the possible histograms characteristics. Even with sophisticated algorithms, there may be some bracketing done (and not for HDR!!!).

As someone else noted in the comments, this is a bit of a "temporary" problem until higher dynamic range sensors come to the market.

Thanks for the article and the clarification on the scene shot, Ctein. I will say that after reading the article, I thought that metering was sorely missing. The bias of the histogram to the right is a maximization of SNR, but not shooting to print -- so the criterion is important: shooting direct-to-print would indicated to me to underexpose, while shooting for post-processing in RAW means to shift the histogram right as much as possible.

cheers!

[Carl Leonardi]

"Carl Leonardi, Not so! With digital you've still got to match the gain setting you've chosen with the exposure settings that match it. If you try to use the aperture and shutter speed settings for ISO 800 when you've got the camera set on ISO 200, or vice versa, you'll get an unusable digital picture 99% of the time."

Mike,
There is a very expensive and sophisticated light meter inside your camera. Using it properly will give you great results. There are so many things the hand held meter does not tell you--like is your lens diaphragm working properly, are your shutter speeds correct, are we photographing something with all one color and no range of contrast. The list goes on. I used to call them (and I had the best meter $$$$ could buy) a close guess meter then, check with Polaroid another almost close guess, then bracket the film because even though the film says ISO 100 it could have been within tolerance of + or - 1/3 of a stop that was color film, B&W was 1/2 a stop + or -.
Learn to read the histogram and you wont go wrong. It's not film, you can shoot lots of test exposures or bracket.

[Mike]

"So, they can change the ISO to a noise-tolerance dial."

Fernando,
No, I don't think so. That's putting it backwards. ISO (or E.I. or REI) is a measure of the sensor's sensitivity to light when it's set on one amplification level. It's not a measure of noise. Noise is just an artifact of high amplification.

Mike

[Carl Leonardi]

"So, they can change the ISO to a noise-tolerance dial."

Fernando,

Maybe we could call it a "push process" dial. Maybe a dial with a picture of the sun, clouds and moon on it. The less light you have the more you will have to open your lens, slow your shutter speed down or increase the amplification of the sensor. You could also increase the light by using flash or other source of illumination. It's all relative to your knowledge of photography and the end results you envision and want to achieve.
If you are not shooting Raw and using Photoshop CS3(or equivalent program) you probably will never see or use your cameras full potential. CS3 is like gasoline to your camera--you could own the best car in the world, if you don't put gas in it, your not going anywhere.
CS3 is now your digital cameras best friend.

[Ctein]

Dear Folks,

Technical asides:

1) My illo photo does, in fact, have about average 'reflectance;' -- it's not unsually dark or light. It just has very big areas of mainly dark and mainly light. Makes the illo clearer, that's all.

2) ISO problem exists whether you capture RAW or JPEG; it just gets worse with JPEG (FWIW, I capture RAW).

3) We all understand that meters (and cameras) don't see reflectance, right? They see photons reflected from surfaces (the technical term would be "luminances"). In the real world, surface luminance is a combination of the surface diffuse reflectance, the specular coefficient, the directionality and location of the light source relative to the subject surface, and the angle of the surface to the camera.

An "averaging reflected-light meter" should really be called an "integrating luminance meter" if one were being technically correct... but that wouldn't make things more intelligible! So we don't.

4) We're not doing data collection, we are not trying to peak the signal-to-noise ratio, we are not trying to maximize the number of grey levels per zone.

We are trying to collect the most useful photographic information we can.

Do not confuse data with information nor scientific goals with photographic ones.

If it looks like crap, it's crap, even if it's got a great S/N ratio. Remember the words of Mr. Natural.

pax / Ctein

[Olaf Ulrich]

Ctein,

ummmmm ... no, in terms of film (or sensor) sensitivity, ASA is not the predecessor of ISO. An expression like 'ISO 200' is meaningless, strictly speaking. 'ASA 200' would be the correct way to put it. 200 means nothing in ISO; 200/24° does (80,000,000,000,000,000,000/200° does, too).

ASA 50 = 18 DIN = ISO 50/18°
ASA 100 = 21 DIN = ISO 100/21°
ASA 200 = 24 DIN = ISO 200/24°
ASA 400 = 27 DIN = ISO 400/27°

and so on.

It has become a bad habit to consider ISO to be just the same as ASA because camera manufacturers tend to short-cut ISO figures to ASA figures as there is not enough room on those tiny scales on the cameras' dials to print the numbers in full length ... still they incorrectly label the ASA scales as 'ISO.' But they aren't. ISO actually is the combination of ASA and DIN.

-- Olaf

[fernando]

Mike,
My apologies for an unclear statement. I meant to do away with the ISO terminology all together. Yes, many of use have a sense of what it is and how to use it, but my comment was to start fresh.

What is little known to many, at least at this point, judging from questions in forums, is that if I am shooting in low-light, say a concert, then it is better to increase the ISO and get the histogram shift right, than to use a lower ISO with the histogram in the middle. (So much as to there being no clipping.) In the mind of many, including my own for a while, high ISO meant, more noise in the image. For digital cameras, it would be good to remove this notion when it is not true.

In the same way that I can calculate a post-compression SNR, by taking the difference between the original image and subtracting the decompressed image from it, I can get a sense of the mean squared error between the two. The larger this measure, the ore noise introduced. A similar measure can be predicted for an sensor, give the the expected light intensity and pixel amplifier operating characteristics (usually versus temperature).

My thinking is it is possible to have a single knob that tells an algorithm to consider ISO and shutter speed (if Av priority), and provide the combination of both parameters given a tolerable SNR for the conditions at hand (which must allow for the aperture available). This means an awareness from the photographer as to how what SNR means what perceptual level of noise. However, we have to learn something new all the time.

This idea is "simple" signal processing ideas extended to this problem, where instead of a fixed film chemical response to light that we can compensate by knowing the film, is for digital to offer the best optimization for two parameters, given one, that achieves a dialed in SNR, for what seems to be contradictory behaviour in low light to what happens in daylight (in as far as my experience).

So, my vote is to get rid of ISO, and not re-define it by leaving it as an internal optimization jointly with another variable (Tv or Av), given the known behaviour of the sensor's amplifiers.

Carl,
That can be name for it... if most people knew what it meant :)

Ctein,

«We're not doing data collection, we are not trying to peak the signal-to-noise ratio, we are not trying to maximize the number of grey levels per zone.»
To me, it is all about maximization of information, whether image, speech, or the signal coming into your phone. As such, the amount of information coming into a single pixel is related to the SNR of light to noise.

Data are just the values affixed to the measured combination of light intensity and noise. Information is related to capacity (e.g, to capture a scene with a certain fidelity). So I agree.

However, a digital sensor is just that: an instrument. As such, it has a certain system behaviour that is not unfamiliar to other non-camera problems and used in many other fields. How well it performs can be measured in terms of SNR, the problem arises with our eyes and mapping an SNR to what is OK, and what is not.

SNR is not going to help with composition, or photographic capabilities. Agreed. It is about having a wonderful, imperfect, instrument to make those instances when we get photography right and preserve it... and to control it as best as possible. Otherwise we would always be happy with Holgas.

[Ctein]

Olaf-- Thanks for catching my goof! All I can say in my defense is that I used to know that [grin].

Fernando-- Sorry, but no; maximizing SNR is not the same as maximizing photographic information. My left-side illo has a better SNR in every pixel than the right-side one... but about 15% of the pixels have one or more channels blown out to value 255. They have great SNR and their tonality is 0, as they can portray one and only on grey level, regardless of the tonal variations in the original scene. Their information content is severely compromised.

pax / Ctein

[Nicolas]

Just a side comment : as this very sound sensitivity redefinition will push ISO values upwards, would it not be better to re-use the old DIN standard, saying the more readable -and displayable- 51DIN (or 51° (which all the french readers will love, including me even if I prefer Casanis or HB... Ooops, I digress too), or whatever in a logarithmic way!) instead of ASA100000?