The Optimum Digital Exposure

[Ray]

It depends on what one's Expose to the right goal is. If the goal is a right aligned histogram, then there are several routes possible, even postprocessing. If one's goal is to optimize overall image quality/dynamic range, then increasing  the number of Photons is the path.

Not really, Bart. An exposure is an exposure and is something that takes place only once at the time the shot is taken. No amount of post processing can change the exposure. We haven't invented time travel yet. 

The concept of ETTR, regardless of the number of photons captured, and regardless of the ISO setting used, implies that the histogram should be aligned to the right to achieve the minimum amount of noise in relation to the other considerations that always apply when taking a shot, such as lighting conditions, subject movement and depth of field.

If one's goal is to optimise overall image quality, then increasing the number of photons captured is not necessarily the path. It might result in a blurry image due to the use of a shutter speed which is too slow, for example.

[BernardLanguillier]

Of course.  Remember we were just guessing why camera manufactures don't build this function into the camera.  If only 1% of the skilled customers need this function, it probably won't make it into the final product, especially if the function is tricky to implement or confusing.

How about the highlight spot exposure capability of the D810?

Cheers,
Bernard

[Fine_Art]

Not really, Bart. An exposure is an exposure and is something that takes place only once at the time the shot is taken. No amount of post processing can change the exposure. We haven't invented time travel yet. 

The concept of ETTR, regardless of the number of photons captured, and regardless of the ISO setting used, implies that the histogram should be aligned to the right to achieve the minimum amount of noise in relation to the other considerations that always apply when taking a shot, such as lighting conditions, subject movement and depth of field.

If one's goal is to optimise overall image quality, then increasing the number of photons captured is not necessarily the path. It might result in a blurry image due to the use of a shutter speed which is too slow, for example.

Lets put the sequence down (manual). The photographer takes the shot based on limits of A and E, which are desired depth of field and motion blur. The sensor has an unchangeable full well capacity of photons.  That is converted to a charge which the A/D converters convert to a closest number. 

ETTR is about letting the raw file use the largest numbers, which makes the whole image have the finest gradations. It is not about the original number of photons exposed. You can use a gain (higher ISO) to get to those finer bins.

[dreed]

How about the highlight spot exposure capability of the D810?

This and a similar feature in Canon DSLRs results in a different tone curve being used to render the JPEG from raw. It has no impact on the way in which raw data (or the exposure) is made.

[dreed]

Of course.  Remember we were just guessing why camera manufactures don't build this function into the camera.  If only 1% of the skilled customers need this function, it probably won't make it into the final product, especially if the function is tricky to implement or confusing.

Then how do you justify the presence of AFMA (auto-focus micro adjust) settings? These are even harder to use properly than any auto-ETTR would.

[Jonathan Ratzlaff]

This is an interesting article and discussion.  For those of us who grew up shooting film there is an analogy. The standard metering on the camera is analogous to shooting transparency film, where the fear was blowing out highlights and letting the shadows fall where they will.  While shooting negative film we used to expose for the shadows because the exposure latitude of the film would hold the highlights.  The problem there was the narrow latitude of the printing paper.  However we used to overexpose negative film by one or two stops as a matter of course when doing weddings to ensure proper shadow detail..
Now with the high dynamic range of the sensor we again have to go back to exposing for shadows.  The empirical part is learing how the sensor handles the highlights and how much we can force it. It's akin to learning how a given film behaves.

The good part is that we no longer have to deal with the constraints of photo printing paper as we can make the adjustments later.  Once again, Jeff's image of Niagara Falls was the first example of this that I ran across.

Now all I have to do is start applying it.

[bjanes]

Not really, Bart. An exposure is an exposure and is something that takes place only once at the time the shot is taken. No amount of post processing can change the exposure. We haven't invented time travel yet. 

The concept of ETTR, regardless of the number of photons captured, and regardless of the ISO setting used, implies that the histogram should be aligned to the right to achieve the minimum amount of noise in relation to the other considerations that always apply when taking a shot, such as lighting conditions, subject movement and depth of field.

If one's goal is to optimise overall image quality, then increasing the number of photons captured is not necessarily the path. It might result in a blurry image due to the use of a shutter speed which is too slow, for example.

Ray,

I think that is not quite correct regarding the histogram to the right. Let's assume that you are shooting with a so called ISOless camera like the D800 in dim light. You would select an aperture small enough for adequate depth of field and a high enough shutter speed to freeze action. Once you have selected the exposure as determined by f/stop and shutter speed, you could take the shot at base ISO. In this case, the histogram will be far to the left. You then increase "exposure" in the raw converter. Its called exposure in ACR or LR, but it is really increasing the gain. Or you could increase the ISO on the camera to move the histogram to the right and take the shot with the same shutter speed and f/stop. The final result would be the same, except that with the higher ISO you have less highlight headroom.

Bill

[BernardLanguillier]

This and a similar feature in Canon DSLRs results in a different tone curve being used to render the JPEG from raw. It has no impact on the way in which raw data (or the exposure) is made.

I don't think that this is correct. The highlight priority mode in the D810 attempts to preserve highlight clipping in the area where the spot metering is performed.

So it does have an impact on exposure, not just about the way the image is rendered.

Cheers,
Bernard

[BobD]

Now with the high dynamic range of the sensor we again have to go back to exposing for shadows...

"Now with the high dynamic range of the sensor we again have to go back to exposing for shadows...[/i]]"
OK, call it macaroni...so lets expose for the shadows.  And how do we do that in digital photograph?... by placing the the brightest desired value in our scene at the [Optimum White Point]... 99+% in our software!

There - "Macaroni". And our shadows are not placed but are dragged and fall on the chip based on the dynamic range of the scene.
If that DR is greater then our sensor's DR then we need HDR... a option we got to much too quickly if you ask me!

Maybe we should stop calling it ETTR - to generic and more a description than a measurement.  Let's call it what it is - the Optimum White Point which is a measurement... 99+% brightness in your camera raw processing software.

(Forgive me... I keep slipping back to my analytical days at Polaroid where our experiment had to be proved then defined in procedures for other to repeat the result.)

[Bart_van_der_Wolf]

Not really, Bart. An exposure is an exposure and is something that takes place only once at the time the shot is taken. No amount of post processing can change the exposure.

Well Ray, that's what the ISO control does as well, it changes the number of exposure photons into a different ADU or DR (Analog to Digital Unit, Digital number), AKA "gain" control. The exposure is unchanged.

The ISO (gain) setting will determine if e.g. each 4 photons will change the resulting ADU in the Raw data by 1 unit (ISO 100), or 1 photon will change the ADU by 1 unit (ISO 400). The exposure (in photons) was the same, the conversion rate was changed. The same can be done in software after the fact. It won't be exactly the same depending on the electronics involved (which can also add noise), or different moments of signal amplification can be used (analog or digital), etc., but the effect will be approx. the same in many cases. There still may be a small ISO/gain increase benefit up to, say, ISO 800. However, it does limit Dynamic range and potentially clips highlights. Doing it in software doesn't risk clipping the highlights.

Cheers,
Bart

[Guillermo Luijk]

Boosting ISO / gain has some merits, but will not produce the same high level of Dynamic range.

In the situation ISO would be used to achieve ETTR that is not correct Bart. ISO should be used for ETTR only when aperture/shutter (photons) don't suffice. In that situation at base ISO you have empty levels on the right side of the histogram. Imagine you have one empty stop for a certain situation; pushing ISO from ISO100 to ISO200 will make us GAIN (not loose), nearly one extra stop of Dynamic Range on Canon cameras.

BTW exposure is about collecting photons... why then software developers call 'Exposure' slider something that only affects digital RAW  values?. I don't mean I don't agree exposure is only about light, just to show that digital photography has stablished new paradigms in photography that need to be reinterpreted. A sensor is not a film anymore and ISO is not a sensitivity anymore.

[Guillermo Luijk]

However, academics aside, when you collect as many photons via ETTR as you can on the sensor, your data is increased - common sense.
The impact is easy to understand - More input data = More RAW converter data for image processing. The result I SEE is less shadow noise!

You confuse 'more data' with 'more information' which is the important thing in practice. The advantage of ETTR is not having more data but having more information through SNR improvement. A good example of more data but no more information (no SNR improvement) is pushing ISO on an underexposed shot on a Sony sensor. Or in nearly any sensor beyond ISO1600, because you get more RAW levels but the same SNR so there is no practical advantage.

Emil explains this fantastically in the article that produced your headache. A simple test to illustrate:



Image on the left has 4 times more data but the same information and robustness against postprocessing. The reason is noise dithering, the same process that takes place in RAW files.

[jjj]

Here's a thought. Rather than arguing semantics and theoretical stuff, why don't people show more photos that demonstrate their point of view as Guilermo + Erik usefully did.

There's also another angle on this. Does increased data etc make for a better looking photograph as opposed to a an image that is technically better?
I recall trying ETTR some years back and whilst in theory the image may have been 'better' I prefered the rendition of the 'lower quality' images.
When I have some time I shall retest, as I have different cameras now and LR/ACR has changed considerably too.

[l_d_allan]

The developers at "Magic Lantern" have included in their arsenal of addons a mode where you can activate an "Automatic ETTR" that uses a histogram built from raw data to calculate the exposure.

From my experience, this works exactly as advertised.

Agree. My observation is that use of ML's A-ETTR provides results that are very close to what is advocated in this article. There are parameters to take into account scenes that have negligible to large amounts of spectral highlights that are allowed to be blown out. The ML design seems to closely cooperate with ACR/LR's PV2012 to have a parameter to allow clipping in one channel, or the Green channel(s).

I'd appreciate feedback from the author or experienced practitioners of what the author is describing / advocating, who have Canon DSLR's with ML installed, on how close ML's A-ETTR implements the article's "best practice" for optimal exposure.

I think of ML's A-ETTR as "OneZone Method #3". Or not?

As others have mentioned, I've found RawDigger to be really helpful to better understand what is going on with the sensor, without all the adjustment "behind your back" that PV2012 does. FastRawViewer from the same developer is also very useful, although simplified with less capability.

Unfortunately it is only available for Canon's EOS line of DSLRs.

Also, agree. ML is perhaps the biggest reason I will stay with Canon DSLR's rather than switching to Nikon or Sony. Their brilliant Dual-ISO really helps with dynamic range, although at the cost of some resolution and additional post processing.

[jjj]

Surely, the sort of person who would use/appreciate ETTR would be shooting manually to get the desired result.
Using auto to get something that needs to be done so very precisely seems contradictory.

[Ray]

There's also another angle on this. Does increased data etc make for a better looking photograph as opposed to a an image that is technically better?

You should know by now, Jeremy, that all appearances such as 'good-looking', 'ugly', 'beautiful', and so on, exist only in the mind of the beholder.

Technically better is unequivocally technically better, provided the science is sound.

There's no accounting for taste. 

[Bart_van_der_Wolf]

In the situation ISO would be used to achieve ETTR that is not correct Bart. ISO should be used for ETTR only when aperture/shutter (photons) don't suffice. In that situation at base ISO you have empty levels on the right side of the histogram. Imagine you have one empty stop for a certain situation; pushing ISO from ISO100 to ISO200 will make us GAIN (not loose), nearly one extra stop of Dynamic Range on Canon cameras.

Dear Guillermo,

Please show me were on this respected website the gain in DR is visible..., I see only decline (see attachement). Of course, Dynamic range is specified as total range, not when underexposing images (which is where S/N ratios for specific luminance levels matter). As I've said before, and we agree on that (I think), with underexposure there are some benefits to (modest ISO increases, more effective on some cameras than others).

Cheers,
Bart

[Guillermo Luijk]

Dear Guillermo,

Please show me were on this respected website the gain in DR is visible..., I see only decline (see attachement). Of course, Dynamic range is specified as total range, not when underexposing images (which is where S/N ratios for specific luminance levels matter). As I've said before, and we agree on that (I think), with underexposure there are some benefits to (modest ISO increases, more effective on some cameras than others).

Cheers,
Bart

Bart, this really surprises me coming from you: one thing is the MAXIMUM DR that a sensor could capture for a given ISO value, which DxO measurements represent perfectly and is always decreasing with increasing ISO.

Another story is the particular situation we have here with ETTR: the RAW data are underexposed at base ISO (right side of the histogram empty, you forgot to red highlight that important point in your quote) and we cannot collect more photons with shutter/aperture. In this situation, pushing ISO:
1. Doesn’t make use loose any highlight information because the histogram is empty there, we are just getting closer to ETTR
2. Provides better SNR in the shadows
If we don’t loose HL DR, but gain shadows DR, we are gaining total CAPTURED DR.

MAXIMUM DR is not always the same as CAPTURED DR, they are only the same when perfect ETTR is achieved.

[Bart_van_der_Wolf]

You should know by now, Jeremy, that all appearances such as 'good-looking', 'ugly', 'beautiful', and so on, exist only in the mind of the beholder.

Indeed, when did excessive noise ever interfere with our ability to make significant tonal adjustments ...,  Oh wait, it does (which explains many of those noise ridden 'creative' images/masterpieces). Maybe that's why ACR has a masking option in its Detail/sharpening panel, hmm.

There's no accounting for taste.  

Apparently.

Cheers,
Bart

[Bart_van_der_Wolf]

MAXIMUM DR is not always the same as CAPTURED DR

Correct, but you didn't specify that:

pushing ISO from ISO100 to ISO200 will make us GAIN (not loose), nearly one extra stop of Dynamic Range on Canon cameras

By only mentioning Dynamic range, you cause confusion amongst some readers. You should, IMHO of course, rather call it improved S/N ratio in the shadows (or "underexposed DR"  to make the distinction clear). That might indeed lead to an improvement of the remaining underexposed dynamic range because the noise floor is lowered, for some cameras more than others. Context is important, so it's better to be specific than to assume that people read the entire post. And even then, I may have misinterpreted your " Imagine you have one empty stop for a certain situation;" premise.

Once photon shot noise dominates (due to higher exposure levels), we'll have a better starting point to optimize the technical image quality than by underexposing and boosting ISO. Of course we sometimes do not have the luxury of choice to achieve both, optimal shutterspeed/aperture and enough light to get ETTR exposure levels. That's why people sometimes use additional light-sources or reflectors if possible.

I think we are essentially saying the same thing, but language backgrounds and trying to drive a certain point home, may interfere a bit.

Cheers,
Bart