Conventional wisdom is that for digital cameras it is best to expose so that the right side of the histogram is as far to the right as possible without blowing the highlights.
However; the raw converter I use allows exposure compensation of +/- 2 stops.
If I under expose does this mean I can bring out detail in the shadow areas or is this a false move. Having preached the conventional wisdom I was asked this question during a talk and I could not answer. Can anyone help.

You want to place as much data to the right so that you end up with the most data in the last stop of the tone curve (shadows). If you have a 12 bit file that can produce 6 stops, the first half of the data is contained in the first stop of exposure data (2048 levels). The last stop has only 64. See:

http://www.ppmag.com/reviews/200612_rodneycm.pdf

I'm not sure I understand your question. Do you wish to underexpose, with an idea of "exposing to the left" to bring out shadow detail?
If this is the question, then no.
You gain shadow detail by exposing to the right (and loose shutterspeed, not to be forgotten!) and then correcting the exposure in the RAW converter.

IMHO, conventional wisdom is correct. At base ISO you might get away with two stops underexposure, since it would be like shooting ISO 400 rather than the base 100. Signal to noise would be worse, especially in the shadows and tonality could suffer. At high ISO, noise would probably be objectionable, depending on the camera.

Actually noise performance is largely determined by the amount of exposure, not the camera ISO. Above unity gain of the camera (which varies from 800 -1600 with most DLSRs--see Roger Clark), if you are strapped for exposure by shutter speed or f/stop necessities, it does not help to raise the camera ISO any further than that of unity gain, and you can increase the exposure in the raw converter.

Two stops overexposure would most likely blow highlights beyond recovery. Raw converters such as ACR can recover 0.5 to 1 f/stops. It really pays to expose properly to the right, just short of highlight clipping.

Bill

If you're "shooting to the right" in a natural lighting situation by using the camera histogram, you're probably (although not certainly) losing specular highlights -- the upper 0.05% of your image -- that help the scene come alive. I prefer to underexpose significantly (a full stop) to capture these highlights (gaining shutter speed) and relying on post processing to correct exposure in a non-linear space. The technique preserves ultrahighlight detail while pulling deep shadows out of the noise floor. I lose a little colour accuracy in zone 1, which I feel an acceptable compromise for the liveliness of the resulting images.

Situations with sky, clouds and/or sun further benefit from this tactic, which mimics the exposure latitude available to analogue media. Again -- yes, you lose some precision in the darkest shadows, but highlights are to my eye a far more valuable resource in creating a natural image.

I expect this is one of those religious flame war topics, but it's one I feel pretty strongly about -- so strongly, in fact, that I developed software to manage noise-reduced shadow recovery when using this tactic.

That's a good reason to expose to the right, but an even better reason is to gain better signal to noise. Shot noise is the major contributor to noise with most cameras and the signal to noise ratio varies as the square root of exposure (the number of electrons captured by the sensor).

The human eye can distinguish only about 70 levels of the 2048 levels in the brightest f/stop of the exposure (Weber-Fechner law, see Norman Koren), so most of those levels are wasted. However, sometimes they come in handy when extensive tonal manipulation is needed.

Bill

If you are interested in the specular highlights, then you should test your camera's histogram and blinking highlights under these conditions rather than systematically underexposing. Many cameras indicate highlight clipping a full stop before it actually occurs and if you expose a stop below the indicated loss of highlights, you could actually be underexposing by 2 stops.

The Nikon D70 was particularly conservative about highlight placement with its metering and many users complained about "underexposure". With my D200, I have learned that I can pretty much trust these indicators and rarely have blown highlights that were not indicated.

With motion picture film and transparencies, it is common to place the specular highlights at 200% so that they really sparkle, but prints lack sufficient dynamic range for this luxury.

Bill

Absolutely.

When I get some time, I really need to play more with this Sekonic and the target they supply to see if it really is a solution for metering 'for the right'. Anyone else have it and tried the calibration with Expose Right in mind?

I think my question is misunderstood. I personnaly always shoot to the right using RAW and do not underexpose. However if I do happen to underexpose, it appears that I can recover the situation using the exposure compensation facilities available in the RAW converter.
If this is an acceptable practice then why promolgate the view that you should shoot to the right?
I have read your responses with great interest and look forward to more.

I've got some brackets at high ISO of this Sekonic target (DNG's and rendered examples) on my iDisk if anyone wants to look em over and comment. Using ACR/LR to (as Michael calls it) Normalize the 'over exposed' image you can see the effects on noise compared to the 'normal' exposure. Look at the noise!

The folder is called Expose to the Right.

My public iDisk:

thedigitaldog

Name (lower case) public
Password (lower case) public

Public folder Password is "public" (note the first letter is NOT capitalized).

To go there via a web browser, use this URL:

http://idisk.mac.com/thedigitaldog-Public

The point is, you're not getting the same data. You're making the image appear lighter at an expense of data in the shadows, or shadow detail if you will. There's no free lunch here. When you shot film and over exposed and under developed (chrome) it altered the rendering of the image and in some cases (depending on how far off you are) affected image quality. Its really the same here. We're talking about correct expsoure to contain as much usable data within the full linear encoded capture.

Under exposure isn't an acceptable practice IF your goal is to capture as much usable data as possible.

I don't expect to accurately represent specular highlights on a print. I expect to have usable detail there that would otherwise wash to a paper white blotch. Even if it's all compressed down to 240-255 there's still good stuff there that makes an image look better to me. And you might be surprised what the right algorithm can do for shadows.

I shoot digital at least a stop down so I have the luxury of highlight data that would otherwise be forever gone. It's my shoulder. And maybe my toe has a bit more noise. But I'm cool with that. The right algorithm can take care of most of it anyway. Highlight recovery is never more than a sop; shadow recovery can do much more.

Why not? You're not supposed to blow out highlight data with the correct exposure. The correct exposure is about capturing the highlight data below clipping but as close to that as possible.



Its only gone if you expose improperly, no one is suggesting that!



If you look at how a sensor, which is just a photon counter captures this linear data, you'll see you're not doing yourself any good and some substantial harm. But its your data so by all means.

What is being discussed here is pretty simple to back up both scientifically (mathematically) and visually using actual image examples.

When I did darkroom work at school, there were plenty of others in the darkroom that didn't expose the paper correctly so they just left the paper in the first develoepr a lot longer or tried to rub areas with their fingers or blew on the print. Heck, if it works for you, go for it. But this isnt' really best practices.

The best thing to be said for under exposure is it makes the preview on the LCD of your camera look better.

I would rather have an exposure that is predominantly dark but has a complete record of the scene than have one that is easy to process but loses highlight detail. With such an exposure I can create a print I'm pleased with.

I think we differ in our definition of highlights, and that's cool.

I've been playing with this concept using my Pentax K100D's in-camera histogram to study how the histogram changes under different exposures of the same scene in the highlite region. Haven't tackled RAW yet.

As I expose toward the right in a scene with say for example a blue sky fading to white toward the horizon, the peak in the highlite region of the histogram will start to gain spikes the closer I make this region of detail brighter through exposure moving the peak closer to clipping. The more I underexpose from this point this highlite peak starts to lose its spikes as it moves farther back from the clipping point. The nice thing is the two shots show very little difference in overall luminance between each other viewed in PS and the shadow regions don't clip to black in either one as well.

If you shoot one stop down, you are losing one stop of dynamic range. With a high dynamic range subject, you may be clipping the shadows and losing data. However, if the camera is able to capture the entire dynamic range of the scene (as shown by a histogram that does not occupy the entire scale), shooting one stop down would only cause slightly increased noise in the shadows, which could be handled, especially if you are shooting with a low noise camera such as the Canon 5D. With a P&S camera, I doubt that this luxury would be worth the cost in noise.

I wouldn't underestimate the utility of highlight recovery in ACR. Specular highlights are usually towards white, have little detail, and can be recovered well. Under such conditions with daylight white balance, the green channel may be blown but the red and blue channels may contain good data, permitting ACR to do a good reconstruction.

I don't think there is any major heresy here as I infer that you believe in shooting to the right as much as possible while still leaving some headroom.

Bill

I note that the Canon 1dMkIII does what I describe above when "Highlight Tone Priority" is selected. Exposes a stop down (ISO 100->200), then pulls the main body of the exposure up whilst compressing the recorded highlights. It's nice that the camera does the work for you, though.

I downloaded the files and did an analysis, which demonstrates good and bad effects of the test series and makes for interesting discussion.

It's nice to have a fancy light meter like Andrew's which gives real time readings, but by looking at the raw files produced by the camera we can also get equivalent exposure data, since the sensor is linear over most of its range. The following analysis involves only the green channel, but could be extended to the others.

Here is a curve from the two exposures spliced together showing the relative exposures of the patches and the resulting values in the raw file which can be decoded with DCRaw, a freeware program much used by digital tinkerers. The raw values are actually 0..4095 but they have been converted to 16 bit format by DCRaw in order to output them by multiplying them by 16.

Click to view attachment

The exposure values on the left are all bunched and hard to read, but they can be spread out with a log-log plot which is standard for plotting characteristic curves of film. The pixel value is normalized to one by dividing by 65635 for a 16 bit file (erroneously labeled 65625 on the image). This notation is more confusing at first, but best once you get used to it. Norman Koren uses this format in his Imatest charts. I used 2 base logs for the exposure, so the values correspond to f/stops.

Click to view attachment

This curve shows that the brightest patch on the f/16 1/50 sec shot is blown and not on the linear portion of the graph.

Now we can look at the characteristic curves of the rendered images. The f/16 @ 1/50 exposure with default rendering has a blown highlight, but this is largely restored by the normalized rendering (shown in yellow). This normalized rendering is similar to that of the normal exposure of f/15 @ 1/200 sec, but the brightest patch is at the maximum pixel value and still slightly blown as shown below. The midtones match fairly well but have slightly different density and slope.

Click to view attachment

Here is the histogram from the blown brightest patch mentioned above. Note that the right side of the bell shaped curve is truncated. This histogram is from the free ware program ImageJ, which does 16 bit histograms, unlike Photoshop.

Click to view attachment

Now finally for the noise, which is measured as the standard deviation of the pixel values in the patches. Actually some of this variation is in the target and possibly in nonuniform illumination, but most of it is likely random noise from the camera. Such noise is primarily photon sampling noise (shot noise), but read noise enters into the equation at low exposure values (see Roger Clark for explanation and a better way to measure noise). This noise is proportional to the square root of the number of photons captured by the sensor and is shown in this plot from the raw data. The noise is actually higher in the file with more exposure and worse in the highlights, perhaps contrary to conventional wisdom which associates noise with the shadows:

Click to view attachment

However, what we are interested in more is the signal to noise ratio. This varies with the square root of the number of photons captured and is greater in the highlights because it is related to the number of photons actually captured (signal) to the square root of the number of photons captured (noise), which reduces mathematically to the square root of the number of photons captured [N/sqrt(n) = sqrt (n)].

Click to view attachment

So in the final analysis, we quadrupled the exposure, captured 4 times as many photons and the S:N improved by a factor of 2 (square root of 4), as predicted by theory. However, the highlights were blown and the recovery was less than perfect. All in all, I thought this was a good blend of theory and practice and worth the effort needed to write it up.

Bill

Well, I have found the replies very interesting. This is one area that I have been playing about with a little over recent weeks.

With my D2x and nx I find I have just over half a stop of "sensible" highlight recovery. By that I mean, reasonably good detail within clouds etc.

I found by moving the histogram to the right by adjusting the exposure, to the point of allowing highlights to clip, and then overexposing by half a stop provided more details in the shady leaves of the test scene and less noise in some other shadow items.

However when using the histogram one must remember how any WB settings MAY effect it!!!

I often find in my images that I am "highlight limited" and personally prefere expose for the highlights and then to push the curves fairly hard at the expense of noise and some shadow detail in order to retain bright details when presented with that situation. (with the exception of skin tones which in colour do not seem to respond well to much pushing. But I certainly don't want to waste space and data at the "high" end and so knowing how much data is there (and slightly more if going to make a composite blending or single shot HDR) or overexposing scenes with limited DR seems to be proving useful.

As for the question, how do I get more than 2 stops of ev correction on post. If using nx, the design folk may well have decided that they feel that is a sensible limit, capture one proved, I think 2 1/5. Bibble seeoms to get a lot of good comments about highlight recovery.

No comments on the analysis?

Bill

Sorry, been on the road (but that doesn't stop others from commenting).

Now that this trip is almost over (writing from lovely Ohare airport), I hope to spend some time looking into this a lot more, talking with the product manager for the new meter and doing many more tests. Got to get to the bottom of this.

This is very interesting thread. There is some agreement that one should expose to the right, but not overdo it. I think that is the problem. The histogram is great tool. But as it is derived from the incamera jpeg, it is not very accurate. With the rgb histogram on the Canon 5D, I have a better idea of what is going on, but in some images there are some highlight values that have poor representation in the histogram. You can think of the histogram as the fraction of pixels with intensity between I and I+DI, where I is the intensity value (0-255). There can be a highlight "toe" that is not well defined in the histogram because the fraction of pixels with that value is low. It can be hard to see the toe under contrasty review conditions. It is only later, when we examine the raw image and set the white balance that we really understand if we got "the ideal" exposure. Those that espouse moderation know what they are talking about. Thiis is an area where experience really counts.

This is very consistent with my experience with a Hasselblad H3D-39. The histogram on the back does not show any evidence of highlight clipping in many images, but it is there when the file is opened in ACR or Flexcolor, and I really dislike the missing highlight detail. Much prefer black shadows to highlights with no detail. It seems that the highlight clipping display on the LCD is much more reliable in showing clipping at the far ends of the histogram and I am starting to rely on that as an important check on the accuracy of the histogram.

This is an oft-quoted model of what is going on, but it is not the most accurate one. The number of levels in a stop is not an issue with current cameras. They have too much noise to be limited by posterization, except for a few cameras at their lowest ISO (Pentax K10D at ISO 100, for example), and then, just barely, and in the deepest shadows.

The signal-to-noise ratio is what matters, in the absence of any real posterization threat. In the deepest shadows, dominated by read noises, the SNR doubles with each doubling of exposure (+1 EC). In the midtone and highlight areas, SNR doubles with each quadrupling of exposure (+2 EC), and the in-between zones, in-between. The benefit is strongest in the dark shadow areas.

The concept of unity gain, as proposed by Roger, is meaningless, IMO. ADC units are arbitrary except in their ability to posterize. He has no evidence for that conclusion; the camera he based it on has no real ISO 3200 with analog gain - it is just 1600 pushed, so of course there is no real noise benefit in using it.

The camera histograms are derived from the JPEG preview, but you do have some control over their appearance via the camera settings. The contrast control, for example, applies an S curve to the data to lower the quarter tones and raise the three quarter tones. This should not affect the end-points of the histogram but does affect the shape towards the extremes. Therefore, many set the camera to low contrast to get a better view of the histograms.

If the camera permits the uploading of a custom tonal response curve, one can use this to calibrate the histogram so that clipping in the histogram correlates with clipping in the raw file. For example, if the histogram indicates clipping when there is none, one can upload a curve with roll off in the highlights. The RGB histograms reflect the status of the channels after white balance, but one can upload a custom WB to obtain an indication of the contents of the channels prior to WB.

Bill

Well, your standpoint does not contradict the shooting to the right idea, as long as you clarify what the right edge of your tonal levels is. If you want specular highlights, then you can expose them as far to the right as possible. "To the right" doesn't necessarily mean positive EC. It just means pushing the tones you wish to preserve just short of clipping. For black spraypaint on on a dark grey wall, ETTR might mean +3 EC. For capturing detail in city lights at night, ETTR might mean -2 EC. The underlying principle is to get the brightest tones you wish to record just short of clipping.

Acceptable and optimum are two different things. If you expose one image a stop more to the right than another, the one exposed more to the right will have usable shadows a stop deeper in real world light, and the acceptable shadows will be good shadows.

Good point! This deserves a lot more attention. It would be nice if the camera makers would just allow us to view a linear encoded Histogram (it will take some getting used to, it's all shoved to one side). Then you need to get the brightness on the LCD way down too.



Exactly. We need to know how to nail the highlights.

Actually, no DR is lost; it is simply shifted from the shadows to the highlights.

Yes, John, we have been over this before. However, it does not make sense to quantify beyond 1 electron = 1 ADU. At that point you have completely quantified the number of electrons that have been captured--you have the actual count and that is all you need.

I have yet to hear your refutation of that point.

Bill

That is assuming that you still have data in all 12 bits of your ADC. If the upper bit is empty, you have lost potential DR. In that case, the maximal:minimal recorded signal drops from 4096:1 to 2048:1 with a 12 bit ADC. With decreased exposure, the noise floor for DR also increases, further limiting actual DR.

Bill

You are talking about the DR of the capture; I tend to think in terms of the medium, as the capture is generally a wild card, as in the specular highlights that the other poster was concerned with.

Yes, in a thread about exposure to the right, are not we all talking about the capture and how to optimize the captured data with due consideration given to the limitations of the medium? Specular highlights and other features of the scene are not really wild cards, but are subject to the laws of physics and scientific analysis.

Many times I have seen people imply that some choice that they make in exposure affects the DR of the camera, and that is what I meant to dispell.



They are wild because you can not accurately measure them in many situations; you can only gamble.

In most cases it is not necessary to render specular highlights accurately, but merely place them above the level of diffuse white in the scene and then let them blow out at higher levels. In movies and slides, the specular highlights are often placed at 200% as discussed in this ICC Paper. In this case you would allow 1 stop of headroom as suggested early in this thread. For prints, probably a bit less headroom would be advisable.

Bill

Nevertheless, it is still up to the individual how much detail they want from these small, brighter areas, and how much they are willing to gamble. There are ways to compress DR locally in images if they want.

I'm pretty certain I have addressed this before. Regardless, the fact is that cameras are *NOT* counting electrons, even if that is what we'd really want them to do. The discreet electrons come packaged in a bundle of analog noise caused by reading, amplifying, transporting, (possibly amplifying again,) and digitizing the electron charge. This extra read noise is *NOT* in units of electrons; it is analog until digitization.

With customized circuitry, Canon has been able to get the level of read noises at the highest ISO down to the equivalent of a few electrons (not a few discreet electrons!). They have been able to do this at the *highest* amplification used in the cameras. Nothing that Roger writes on his website addresses what may or may not happen with more and better amplification; he simply jumps to the conclusion that nothing is gained, and uses the fact that his 1Dmk2 has the same total read noise in electrons at ISO 3200 as it does at ISO 1600. That is not any real support for his conclusion, because ISO 3200 *IS* ISO 1600 on that camera. Had he used a Minolta K7, which uses real amplification at ISO 3200, he would have measured slightly less noise at ISO 3200, and if he had actually looked at the shadows, there would be slightly less line noise at 3200, and less chromatic noise in a RAW at 3200 than 1600 pushed to 3200. I'd offer the 1Dmk3's ISO 3200 as additional support for my claim, but the fact that it is 14 bit may make you feel that the goal post for unity gain has moved (despite the fact that mk3 ISO 3200 quantized to 8 bits is still far less noisy than the mk2's ISO 3200).

I'm sure I have shown you this chart before, in previous refutations of the "unity gain" limit:



That is the total read noise, and the isolated horizontal and vertical line noises.
The total read noise, the yellow line, is scaled to 10% to fit in with the others, and the vertical axis is the read noise normalized to ISO 100 for all other ISOs, as standard deviation in ADUs (which can be considered arbitrary units of electrons). The noises clearly show no sign of flatlining completely by 3200, as far as the trends up to 1600 are concerned, especially the line noises, which are far more visible than their statistical strength suggests. Horizontal line noise *is*, without a doubt, the most troublesome aspect of high-ISO shadow areas in Canon cameras.

Roger has nothing to really support his unity gain hypothesis; he is simply applying the concept of one equals one, but these ones are really apples and oranges; one is discreet integer values, and the other is discreet multiples of a single value, with variance at a finer degree. The ADC in these cameras can *NOT* count electrons. They can only get so close to counting them, and by all appearances, with Canon's technology, the more you amplify the signal, the more you can reduce the inaccuracy, which is why it is illogical to declare that something as arbitrary as the ADU unit is a meaningful limit to practical amplification. And the ADU truly *is* arbitrary when it is fine enough not to cause posterization of RAW data. Only when it is coarse enough to cause posterization does the actual absolute meaning of the ADU have any value (the ability to posterize).

Even without direct support on the camera, it could conceivably be done now with a computer hooked up to the camera; a program could look for new RAW files on the card or in the computer, and display a RAW image and/or histogram on the screen. Not for action shooting, of course!

Modifying the contrast seems like a good idea. But in the final analysis, the camera manufacturers have let us down. There are not enough bins in the histogram and it comes from the jpeg, not the raw. Because there are so few bins in the histogram, its shape is poorly defined. This matters most in the highlights. In some case parts of the histrgram can be distorted becasue the bin-width is too large. One experiment for someone with ambition and too much time is to measure the in-camera histogram and compare its shape to the one in photoshop. It would be good to do this experiment for several lighting conditions

Thanks for sharing this analysis. This whole discussion is helping me to see some of the things I am doing wrong. I think I resist going up to a higher ISO fearing noise, but I end up pushing it up on the computer and maybe coming up with even more noise (I use a 20D primarily)

______________________
"You cannot escape the responsibility of tomorrow by evading it today." Abraham Lincoln
Company – Projects (Affordable Stock Photos – Sell Your Photos)

It is a good idea, but you have to experiemnt with whatever camera you are using. Try different settings of jpeg contrast and see how 'no clipping' or 'moderate clipping' on the camera's LCD screen (flashing red patches)translates to the degree of recoverable highlights in your RAW converter.

I've got my 5D set so that I know a small amount of red flashing in the brightest parts of the sky is recoverable in ACR, but not if a large area of the sky is flashing.

I think that setting is minimum contrast (it's a long time since I set it). At maximum contrast, I'd have a situation where the histogram and blown highlight warning would have the whole sky flashing, causing me to underexpose.

[quote=Ray,Jun 28 2007, 08:16 AM]recoverable highlights in your RAW converter.
[/quote]

"Highlight recovery" is a usual term when talking about RAW developing, but it's most of the times wrong. What you get in ACR pushing down the exposure slider is not any highlight recovering at all. If your previously blown highlights become not blown through underexposure tuning, is because they were actually never blown on the RAW file. It was YOU in the developing process, commonly with the white balance (which implies heavy channel scaling by factors usually greater than 2.0 in linear, i.e. +1EV correction, and easily beyond 2.5, nearly +1.5EV), who blowed those areas.

ACR does not apply any real recovery of highlights. DCRAW in its -H2 to -H9 modes does; DCRAW interpolates truly blown pixels to values close to those in the boundaries; it "invents" colours where they were previously blown.

I just wanted to point this as most people talk about hightlight recorvery when thery are just simply referrering to "no-blowing underexposure".

Here is a sample that demonstrates how white balance can blow image areas that were not in the original RAW file. This image was developed without applying any white balance (hence the greeny colour):




If you look into the window area and compare that result (ANTES) to the one obtained with an ACR type white balance applied (con balance de blancos) you can see how much information YOU are blowing just with your white balance:

[/quote]

I've read this topic over and over and only grasp about a quarter of what you folks are saying. And I keep asking myself, so what! Are you only discussing theory or is there some real world benefit to your debate.

I ask simply because I'm so new to photography and because of my advancing years, don't have time to theorize. If I'm unsure about exposure, I bracket. If I still think that I still have not gotten the whole range, I bracket and merge to HDR.

Please don't think that I'm being disrespectfull, because that is not my intent.

To give you an idea where I'm coming from, my work is for personal enjoyment, don't sell or show. But rest assured, I do have a very demanding audiance.

FWIW

I provided a link of John's post to Roger Clark and invited him to respond, and he was kind enough to provide the following explanation.


Hi Bill,

You can post this response if you wish to the luminous-landscape forum
as I have not registered to post there.

John Sheehy posted a plot of noise versus ISO for a Canon 20D.
The way I read the plot is that the noise at ISO 1600 is the
same as ISO 3200 and the noise is greater at ISO 800.
This is the same as other people's measurements.

The Unity Gain ISO for the 20D is 1200, so one would expect
a slight improvement from ISO 800 to 1600 (ISOs in between
these factor of 2 values are reportedly scaled and not true gains).
Amateur astronomers are pushing these cameras to their limits
and everyone I know has concluded that there is no benefit in
going to ISO 3200 (query the literally thousands of people
on the digital_astro yahoo group). Most are using ISO 1600
and some ISO 800. So I don't see John's plot in conflict with
the Unity Gain implications. I also have never seen a clear demonstration
that you could actually get more out of an image at ISO 3200
versus 1600 on the current suite of 12-bit/pixel cameras.
A number of amateur astronomers have tested this too and
come to this conclusion (that ISO 1600 "gets it all").
In fact you actually lose a stop of dynamic range in going from
ISO 1600 to 3200.

Regarding the 1D Mark III, part of the "read noise" is A/D converter
noise. The use of a 14-bit A/D reduces that noise and Canon reportedly
has a half to one stop better noise floor performance.
More importantly, Canon has reportedly reduced the fixed-
pattern noise (e.g. line noise). That improves the
perception of noise making the higher ISO images look
better. Until these cameras get into the hands of people
who run rigorous tests we really won't know the
implications of ISO, 14-bits, and noise floors.

Roger


The implications for shooting in very dim light are that it does not really make much sense to exceed the unity gain under these conditions. You get no more information and lose one f/stop of dynamic range. The unity gain of the Nikon D200 is 800 according to Roger's tests. According to the unity gain theory, when shooting in dim light with this camera in raw mode, it would be best to set the camera to ISO 800 rather than 1600. If there is enough light to expose at ISO 800, you get better dynamic range and less noise. If "underexposure" occurs at ISO 800, you merely use the exposure control of ACR (or whatever raw converter you are using) to brighten the image.

Bill

He seems to have missed the point.

Perhaps you should ask him to re-read John's post, particularly the part about the ISO modes above 1600 on pre-1D MkIII Canon cameras.

wmchauncey -- Here's the real deal: Current megapixel digital sensors are like film with no shoulder at all. When someone says "this looks digital" they really mean that the highlights are clipped hard. No one exposes film to clip highlights hard; they haven't for decades. I'm not sure why we're supposed to expose our digital sensors that way, even though they have the smoothest response right up near that wall of clipped highlights. So what if they are? Am I going to regret a minor amount of additional more noise by boosting my RAW file if doing so makes a scene look natural? No. I am not, my curators are not, and my customers are not. They will look at my photos and say, "wow."

The dumb thing is that most digicams push the metering right up near that wall so that signal/noise ratios are as low as possible (probably for marketing's sake) -- at the cost of hard clipped highlights. Well, I don't know about you, but I don't sell "signal/noise ratios". I sell pictures. And I've found that I can make the kind of pictures I want by shooting "underexposed" according to the camera, then "push-processing" the resulting RAW file. I get a little bit more noise, but a much more natural looking result.



This is an extreme example. Exposed -2EV, shadows recovered up to 4 stops. Detail visible in the shadowed brush on the left and around the telegraph pole on the right. And the highlights on the rails look natural, and the sky isn't blown out, and it feels just like it felt when I was there -- a bright, clear, early morning.

Remember that Canon has added a "highlight tone priority" mode to the 1DmkIII -- which does exactly what I describe above. If you don't trust me, trust Canon. The era of hard clipped digital highlights is finally coming to an end. It's been a long time coming. And if you don't have a 1DmkIII, you can still do this by "underexposing" in camera, and "push-processing" in RAW.


-s

[edit -- removed unnecessary editorializing -s]

Really? Where is the proof for that?



You get no more information BECAUSE THEY REALLY AREN'T TRUE AMPLIFIED 3200. How many times do I have to repeat this extremely relevant fact? Roger's conclusions, and those of the astronomers he's asked, are all based on ISO 3200s that are really ISO 1600s under-exposed, with the RAW values doubled and almost a stop of hightlights clipped away for no good reason.



What does that have to do with unity gain?

It is true for most Nikon cameras more than a year or two old, because most of the read noise occurs at the initial read on the sensor, and only a clean, low-gain amplifier is used to feed the ADC. Total read noise in electrons is very similar at all ISOs, and only varies because the absolute_signal-to-ADC_noise is different.

The same principle can apply between ISO 200 and 400, as well as 400 and 800 with cameras that do not have lower absolute (electron) noise at higher ISOs (like the Pentax K10D). Nothing special happening at unity gain. Even when the ADC noise makes a difference between ISOs, it makes the most difference between lower ISOs, explaining why there is less loss pushing 800 to 1600 than 200 to 400. The higher the noise before the ADC, the less increase there is in total noise from the ADC, because of the non-linear way in which noise sums.

Only poor circumstantial evidence exists for the unity gain theory. An ADU:electron ratio is only completely relevant if the total read noise is low enough so that no two quantities of electrons are digitized as one. That actually requires far greater than 1:1 with even 0.1 adu of analog read noise, for total accuracy in counting.

As far as 14 bits are concerned, they do nothing for IQ at ISO3200 with the mk3. Truncated to 8 bits, and then converted to RGB, the mk3 has less noise than the 1dmk2 with 12 bits. Even 12 bits at high ISOs is overkill for 99.99% of uses.

Roger's idea of testing this type of thing is to take a linear conversion, and then quantizing it. That is nothing at all like quantizing the RAW data, and then interpolating/demosaicing it and performing WB.

I don't have unlimited time on my hands, but have done part of the experiment that you suggest with a Nikon D200. Illumination was 5000K. I took shots of a Stouffer step wedge, increasing the exposure in 0.33 EV increments until I observed clipping on the camera histogram with contrast set to normal. I then examined the contents of the raw files (converted with DCRaw) and the preview in Adobe Camera Raw. The displayed results are for the last exposure without clipping.

Here is the camera histogram. Note that the head of the wedge (which includes step 1) is just short of clipping and is the large spike towards the right. The small spike to the extreme right is blown background without the base density of the wedge.

Click to view attachment

Here are the pixel values in the raw file (in 8 bit notation). Note that step 2 (down 0.3 EV) is not blown in any channel, whereas the green is in step 1 is near maximum.

Click to view attachment

And this is how ACR views the file:

Click to view attachment

Since the highlights are 255, I used the exposure control to decrease exposure until the value fell below 255, which took place at -0.2:

Click to view attachment

With my particular camera, the camera histogram from the JPEG preview gives an accurate indication of clipping. For most exposures, I like to have the exposure just short of clipping. If the highlights are not critical and I want better shadow detail, then I may allow some highlight clipping and perform recovery in ACR.

Bill

John,

I don't really feel qualified to comment on the fine points of the work that you and Roger have done. At this point, I would look at the qualifications and background of the authors. Roger has a PhD in astrophysics from MIT and is professionally involved in various imaging projects at NASA and has 179 peer reviewed scientific papers (Bio RN Clark). What are your qualifications?

Bill