The Optimum Digital Exposure

[MHMG]

John. You can't "overexpose" a sensor. Once the wells are full the photons fall on the floor. It's a brick wall.

You can't overexpose a sensor, but you can overexpose an image. And when you do, highlight color and tonal detail cannot be recovered from the RAW file because what Michael said, those photons "fell on the floor"

In the "good ole" days of the film era and the classic zone system, the rule for making optimal negatives was to "expose for the shadows and develop for the highlights". It made total sense but begged the use of a spot meter to perform the task consistently under real world lighting not studio conditions. For the modern RAW "digital negative" workflow this classic rule can be stated as just the reverse, i.e.,  "Expose for the highlights and develop for the shadows". "ETTR" is just another way of saying the same thing, and it's also how we had to expose color reversal film except that we had no real control over the development of the shadows with color reversal film. With todays RAW image editors we do.

I agree with Michael and others that the camera engineers could indeed build a decent ETTR mode into the camera metering system, but I suspect part of their reluctance to do so is that it would require a total rethink on how to construct a good looking OOC jpeg from that ETTR RAW data. Still, it's possible, and if anyone is likely to do it, my guess would be Fuji will get there first.

Re: the article:  It makes some good points, but the notion that "A good digital exposure looks as if it has been dipped in skim milk!” will lead to totally clipped highlights with many of todays digital cameras, IMHO.  This exposure advice is probably too simplistic because there are so many camera and RAW editor variables that factor into the determination of the optimal RAW exposure. Neither the camera blinkies nor the clipping warnings in LR are a precise indicator as to when full color and tone is being destroyed in critical highlight values of the scene. Today's cameras' "smart metering modes" like my D810's matrix and highlight metering modes are still trying to play nice with OOC jpeg and "Dlighting" jpeg processing methods. As such they will miss the optimal RAW exposure far too many times to make me comfortable with those modes out in the field. I resort to, wait for it, spot metering mode with liberal use of AE LOCK and then recompose the frame, or as runner up, the center weighted mode. Both of these modes aren't trying to outthink me, but as the "thinking" photographer I have to pay more attention to correct exposure to master those older "traditional" metering modes

Lastly, re the gamut warnings in LR. I find that it is entirely possible to set the RGB model to Prophoto, enable a particular camera profile, and then set the various sliders such that I can make the gamut clipping warnings go away even on images where delicate highlight colors and tones in the RAW file are still truly clipped. I have found that the single best analytical tool to really get a handle on camera-specific RAW image exposure is to use the Xrite color Passport target and pay particular attention to the light pastel color patches in the target that Xrite added to the overall target for white balance corrections. These very light tones of subtle difference in hue provide a very sensitive test for when delicate highlight colors are getting overexposed to the extent that they are impossible to fully recover with "magic" settings in LR or other RAW editors. When the camera sensor is right on the brink of clipping, recovery of those delicate colors may still be partially recoverable but definitely not fully recoverable. For my D810, it occurs with about 1.3 stop extra exposure to reach that point when metering carefully from an 18% reflectance gray card. On my Fuji S3, it takes about 3 stops overexposure to begin to lose those delicate highlight colors in RAW processing. Moral of the story is that ETTR tolerances are highly camera dependent, and a hand-held spot meter is still a wise choice if you are serious about ETTR optimal exposures.

cheers,
Mark
http://www.aardenburg-imaging.com

[michael]

Thanks for your comments Bob.

I'm just waiting for one of the companies to file a patent for it. I have emails going back at least 10 years that would count as "prior art".

Cheers,

Michael

[bjanes]

I agree with Bill, flowers can have amazing saturation thus presenting not only overexposure risk, but some also risk of underexposure, e.g. with near zero Blue and very high Red channel signals (at the same time).

RawTherapee at least offers a solution by allowing to change the per channel pre-demosaicing signal levels, I wish other converters allowed such manual control.

Cheers,
Bart

Bart,

The ability to set per channel WB multipliers in the raw converter (as can be done with RawTherapee, which also has an option to show the raw histogram) would be helpful. Rather than leaving the green multiplier at 1.0 and having red and blue multipliers greater than unity, one could set the multiplier for the most exposed channel to unity, and the other multipliers to less than unity, as Guillermo discusses here for dcraw.

If the raw converter had a true exposure slider affecting all channels linearly, couldn't one accomplish something similar by merely decreasing the exposure. With ACR and LR, the exposure slider with PV2010 seems to work reasonably well for this purpose.

Bill

[dreed]

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.

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

[Bart_van_der_Wolf]

Bart,

The ability to set per channel WB multipliers in the raw converter (as can be done with RawTherapee, which also has an option to show the raw histogram) would be helpful. Rather than leaving the green multiplier at 1.0 and having red and blue multipliers greater than unity, one could set the multiplier for the most exposed channel to unity, and the other multipliers to less than unity, as Guillermo discusses here for dcraw.

Exactly what I'm talking about. Whitebalancing and 'exposure' optimization bundled together. Of course that's harder to do in Raw capture, but then one would need to get flexible exposure per channel (while recording white balance as multipliers). When the data is still in linear gamma space, not really all that hard to do.

If the raw converter had a true exposure slider affecting all channels linearly, couldn't one accomplish something similar by merely decreasing the exposure. With ACR and LR, the exposure slider with PV2010 seems to work reasonably well for this purpose.

Yes, after capture, it's easy enough when still in linear gamma space. Depending on the software, there may be some complications with color profiles ('scene referred' versus 'output referred'), also shown by the PV2010 (and later) (non-)predictability. But then, what's one more challenge after eliminating several others.

Cheers,
Bart

[deejjjaaaa]

I wish other converters allowed such manual control.

and they do, for example RPP or Iridient have per channel multipliers (RPP only and Iridient as an option)... there are many different raw converters in this world

[deejjjaaaa]

but one can obtain a reasonable approximation of the status of the raw RGB channels after white balance in ACR or LR by setting the color space to ProPhotoRGB and using PV2010 with the sliders on the main panel zeroed out (set to 0).

there might be hidden exposure corrections in LR/ACR ... so unless in addition you use (or verify that your particular combination of camera model, profile, iso, etc does not do this) a particular .dcp profile for a particular camera model (and even for a particular ISO there - because Adobe might use different hidden expocorrections for different ISOs) ACR and LR are simply not the right tool to be bothered with

[bjanes]

there might be hidden exposure corrections in LR/ACR ... so unless in addition you use (or verify that your particular combination of camera model, profile, iso, etc does not do this) a particular .dcp profile for a particular camera model (and even for a particular ISO there - because Adobe might use different hidden expocorrections for different ISOs) ACR and LR are simply not the right tool to be bothered with

That is a good point, but with my D800e at base ISO and with the Adobe Standard profile, there appears to be no hidden expo corrections as can be verified by comparing the ACR histogram to the RawDigger histogram. While not ideal for some purposes, ACR/LR is the raw converter that many of us use for routine work.

Bill

[Guillermo Luijk]

Erik, not really I think. ETTR is the same for every sensor. Meaning to gather as much light as possible. ISO and shadow priority is a different matter.

Wrong. See what happens on a Canon sensor when you fiddle with ISO at constant exposure (aperture/shutter), i.e. constant gathered light:



On Sony sensors that improvement is negligible.

ETTR tools:

- Aperture, shutter and ISO for Canon sensors.
- Aperture and shutter for Sony sensors.

The rest will fall somewhere in between.

[deejjjaaaa]

On Sony sensors that improvement is negligible.

A7s..........

[bjanes]

Exactly what I'm talking about. Whitebalancing and 'exposure' optimization bundled together. Of course that's harder to do in Raw capture, but then one would need to get flexible exposure per channel (while recording white balance as multipliers). When the data is still in linear gamma space, not really all that hard to do.

Yes, after capture, it's easy enough when still in linear gamma space. Depending on the software, there may be some complications with color profiles ('scene referred' versus 'output referred'), also shown by the PV2010 (and later) (non-)predictability. But then, what's one more challenge after eliminating several others.

Cheers,
Bart

Bart,

Thanks for your comments. I am posting the results of a recent experiment with white balance for a red flower using the Nikon D800e for information and further comment. The WB was set to Daylight in the camera (at around 1:30 pm, CDT at 42 N lat on 10/25/2015, -5h GMT) with the Nikon standard Picture Control and file space set to AdobeRGB. The wb multipliers for RGGB are 1.973, 1.0, 1.0, 1.375 as reported by RawDigger's EXIF tool.

The metered exposure was 1/100 s @ f/9. The Red channel shows severe clipping.



However, the raw histogram shows underexposure in all channels, with the red channel about 1.5 EV below saturation.



For a proper ETTR exposure, it is necessary to increase the exposure, rather than decreasing it as would be indicated by the red channel histogram on the camera. Doubling of the exposure gives a better ETTR exposure, but the red channel is still 0.5 EV below clipping, which would allow for specular highlights if any were present. The red histogram was shifted 1 EV to the right, as expected with the 1 EV exposure increment.



In ACR with PV2012, a negative exposure of -0.65 EV gives a good histogram. Perhaps PV 2010 would give more accurate color, but I used PV2012 since that is what most photographers would likely be using.



Bill

[BobD]

Bob,

What is your view about the value of the highlight priority metering mode of the D810?

Since it is linked with the spot meter mode of the camera, it seems like a pretty good solution, isn't it?

Cheers,
Bernard

Bernard,
If I'm understanding this feature correctly, it looks like using this function is diametrically opposed to what is needed. It appears as if they are basing the exposure on the highlights which drives the the exposure to the left - underexposing.
"Although spot metering targets a selected area, targeting can be difficult when your subject is in motion. Matrix metering balances exposure over the entire frame, including the background, and may consequently leave highlights overexposed, or "washed out." In contrast, highlight-weighted metering automatically detects and meters highlights for optimal exposure with less washout (and less fiddling with the camera)." As Michael said in an earlier post, "it appears as if they just don't get"!

If you're using the OneZone Method #2: +1.3 EV plus and minus 2/3 EV, I would use the matrix metering.
Using your in camera spot meter can be a little tricky. Here is how I address of spot metering in my book:
"Although you can use your camera’s spot meter, there are a couple of things to consider when doing so:
•   First, the Spot Size: The spot is relatively large and varies in size with focal length of the lens being used. This can be useful if you are using a zoom lens. When you zoom large you narrow the spot meter area and can get a more accurate reading.
Second, the Accuracy of Reading: If your zoom lens does not have a constant f-stop throughout the entire zoom range, zooming larger in order to narrow the spot meter area will produce an inaccurate exposure when you zoom smaller to recompose. (For more information, see more In-Camera Spot Metering Considerations in the Appendix)"

Bob

[BobD]

I agree with Bill, flowers can have amazing saturation thus presenting not only overexposure risk, but some also risk of underexposure, e.g. with near zero Blue and very high Red channel signals (at the same time).
RawTherapee at least offers a solution by allowing to change the per channel pre-demosaicing signal levels, I wish other converters allowed such manual control.

Cheers,
Bart

I also agree that flowers have amazing saturation. However, this can be also remedied in ACR (LR or PS) using the TAT tool to lower the red/increase the blue saturation or by using the Tone Curve by adjusting the red and blue channels.

[Torbjörn Tapani]

Wrong. See what happens on a Canon sensor when you fiddle with ISO at constant exposure (aperture/shutter), i.e. constant gathered light:

[image]

On Sony sensors that improvement is negligible.

ETTR tools:

- Aperture, shutter and ISO for Canon sensors.
- Aperture and shutter for Sony sensors.

The rest will fall somewhere in between.

But if you keep constant exposure (shutter/aperture) then it's not ETTR in the ISO100 example. Am I missing something here?

Was the image ETTR:ed at ISO 100, and then the same exposure made in ISO1600 to produce that result? Then you probably have one of those examples where ETTR would yield lower exposure than metered. Say we have some highlight out of frame that was ETTR:ed for.

But sure. Other than base ISO is good for separation in shadows. This is true even for the near ISO-less D7000 and D800. Which we discussed recently: http://www.luminous-landscape.com/forum/index.php?topic=94383.0

It's absolutely possible I'm flat out wrong but enlighten me.

[Bart_van_der_Wolf]

But if you keep constant exposure (shutter/aperture) then it's not ETTR in the ISO100 example. Am I missing something here?

Correct. All the images show is that for some cameras that increasing the ISO can be beneficial for shadow noise, although it also reduces the dynamic range.

ETTR is all about collecting as many photons as possible, without clipping, to improve Dynamic Range. That's typically low ISO territory.

Cheers,
Bart

[bjanes]

I also agree that flowers have amazing saturation. However, this can be also remedied in ACR (LR or PS) using the TAT tool to lower the red/increase the blue saturation or by using the Tone Curve by adjusting the red and blue channels.

This could be the topic of another thread, but with the image I used as an example, there is no trouble bringing the image into the gamut of ProPhotoRGB with ACR or into the ProPhoto like working space of LR, and I think the best approach would be to optimize image in those spaces. For printing, one would probably not want to eliminate all clipping of the red channel, but to reduce saturation until important image details are not burned out as judged from soft proofing. With LR, one could create a virtual copy for proofing and edit the image in real time. With Photoshop and ACR, one could render into the printer space and perform adjustments there. How would you handle this situation?

Regards,

Bill

[williamchutton]

Since 2011 I followed the succinct advice presented by LuLA member ejmartin (post #88).

http://www.luminous-landscape.com/forum/index.php?topic=56947.msg463071#msg463071

I don't use Canon cameras, so ejmartin's advice has served me well.

Bill Claff's brand specific data on photographic dynamic range shadow improvement vs ISO is interesting.

https://home.comcast.net/%7ENikonD70/Charts/PDR_Shadow.htm

[kwalsh]

Worth mentioning if you are already going to deal with in camera previews that are blown out/over exposed to do ETTR you might as well make those images green and more useful.  Which is to say set a custom WB (sometime called UniWB) that will make your in camera JPEG based histograms more closely match the RAW histograms.  For most cameras this means AWBing on a magentaish target with the end result being green previews.  Depending on the camera the technique to do this can be easy or hard, one shot or iterative.  But in the end you have a custom WB setting that will give you close to RAW histograms.  Google it.

Also I've got to say that writing and encouraging people to go to 99% on SW like LR as a general practice is foolhardy.  Bad things happen near saturation and it can be extremely unforgiving.  A little more shadow noise is easily handled in post processing.  Blown highlights and exciting color errors as a result of non-linearity near saturation capacities of wells is much worse to deal with.  So in that vein I find the "method 2" which is very close to what I use to be much more practical and sensible.  Most people are going to be much better served by being a half stop below "99% optimal" than actually running up against saturation.  Especially when that saturation point is determined rather erroneously from a LR histogram.

The ETTR concept is very sensible.  Going to 99% is introducing new problems without actually improving things in the shadows and mid-tones hardly at all.

[Guillermo Luijk]

But if you keep constant exposure (shutter/aperture) then it's not ETTR in the ISO100 example. Am I missing something here?

Was the image ETTR:ed at ISO 100, and then the same exposure made in ISO1600 to produce that result? Then you probably have one of those examples where ETTR would yield lower exposure than metered. Say we have some highlight out of frame that was ETTR:ed for.

But sure. Other than base ISO is good for separation in shadows. This is true even for the near ISO-less D7000 and D800. Which we discussed recently: http://www.luminous-landscape.com/forum/index.php?topic=94383.0

It's absolutely possible I'm flat out wrong but enlighten me.

What I demonstrated is that ISO is another tool to achieve ETTR on Canon sensors when aperture/shutter don't suffice because of the lighting conditions and DOF/motion blur requirements.

So yes, ETTR is about ISO as well, not only about collecting photons.

Regards

[ErikKaffehr]

Hi,

Guillermo's post illustrates what I was thinking about. The way I see it, ETTR is about maximising the number of electrons captured. But on Canons and some other cameras, increasing ISO will reduce readout noise, so increasing ISO and doing ETTR may make some sense.

On "isoless" cameras I would say that keeping base ISO and exposing as much as possible is the key to achieve maximum image quality.

Best regards
Erik

Wrong. See what happens on a Canon sensor when you fiddle with ISO at constant exposure (aperture/shutter), i.e. constant gathered light:



On Sony sensors that improvement is negligible.

ETTR tools:

- Aperture, shutter and ISO for Canon sensors.
- Aperture and shutter for Sony sensors.

The rest will fall somewhere in between.