7

I'm a beginner photographer and I have read articles that advocate exposing to the right (ETTR) and articles that advocate exposure to the left (ETTL), and this has left me quite confused. I assume this are equivalent to exposing for the shadows (ETTR) and exposing for the highlights (ETTL). Is this correct?

ETTR is advocated so that you expose enough until you don't clip highlights, so that you can get less signal to noise ratio in the shadows. ETTL is advocated because the shadows preserve better details than highlights. Are these two complementary or is the use dependent on application? I'm quite confused as to what to use or what not to use and when.

3

This may entirely depend on what camera you use. Canon cameras are rather notorious these days for not having competitive dynamic range. They have decent dynamic range, but it is no longer competitive, and there are alternative options that allow you to preserve considerably more shadow detail without the need to either worry about using ETTR at all, or you simply don't have to be as aggressive with ETTR.

For a full explanation on dynamic range, which is a bit of a prerequisite for understanding why ETTR is needed, I've written a comparison article on my site: Dynamic Range

For a quick example, here is an A7r (left) and 5D III (right), both exposed for the same scene such that the brightest highlights were just barely clipped, thus making complete use of the entire dynamic range of both cameras:

enter image description here (See link above for higher resolution)

The 5D III is already showing more noise with a mere 3-stop push. The A7r holds up well until a full five stop push. I even went "extreme" and pushed the images well beyond five stops (about seven stops for the shadows, with the additional adjustments):

enter image description here

Again, the A7r holds up quite well, the 5D III just doesn't cut it. (Newer Canon cameras have less banding, however fundamentally their dynamic range has not improved much...maybe a third of a stop. You won't see much better performance with a 7D II, a 6D, nor even a 5Ds.)

For reference, here is what the original images looked like strait out of camera (as rendered by Lightroom):

enter image description here

Lot of deep shadows there, some that appear dead black. LOT of information that can be recovered.

A lot of people complain that having more dynamic range in the camera just means you have lower contrast in your images after you lift the shadows. That is a naive reaction, usually one given due to either distinct and strongly held brand loyalties, or simply out of naivete or a lax understanding of post processing. A shadow push is simply the beginning step of a processing workflow. Contrast can be restored to the image post-push:

enter image description here

While you can see some noise in the 7-stop lifted shadows of the A7r here (top), and that 7-stop lift does wash the image out...you should never just stop there. As you can see, the same image with some additional adjustments to restore contrast (bottom), and a little bit of NR, result in very clean shadows, despite the fact that they were buried seven stops deep in the blackness of the original image.

The 5D III, however...despite similar edits, and much heavier NR, it simply doesn't hold up. With the 5D III, despite pushing ETTR to the limits, I was only able to preserve so much shadow detail. Now, without ETTR, I would have lost EVEN MORE shadow detail with the 5D III...and I would have lost nearly as much with a 6D, a 7D II, or a 5Ds (despite all of them having lower dark current and less banding, their read noise is just about as high.) ETTR becomes essential with a Canon camera. If the shadows are very important, clipping highlights even more may be acceptable. With a still scene, HDR would probably be preferable.


Before I can really answer your question about ETTR, my first question is:

What camera do you have?

If you have a Canon camera, then you do need to pay much closer attention to what range of tones you are preserving in your photos. Technically speaking, "correct" ETTR means you NEVER throw away any highlights. There are various schools of thought on that, and in some cases, throwing away some highlights may be acceptable, and if there is shadow detail that is more important than those highlights, the simple logical conclusion is to throw away the highlights. For purists, that would be going beyond ETTR, but in general, I think you do what you have to in order to produce the photo you want...and if that means throwing away highlights, and you are consciously making that decision, then its an entirely acceptable option.

Is there such a thing as ETTL? Personally, I don't think ETTL is a viable concept. We are talking about signals here...to be more specific, we are talking about digital signals and SNR. In a digital signal that is produced by an image sensor, you have a pretty hard cutoff at the point where highlights clip. This is due to the nature of how the signal in each pixel is read out and converted to a digital number. The analog response of the sensor itself may not be linear throughout the entire range of it's literal sensitivity, however most camera manufacturers limit the "usable" range of charge to within the largely linear response, and any signal beyond the range of linear response at the high end is simply clipped to white.

When it comes to shadows...fundamentally, shadows are not clipped, not by any natural consequence of sensor design. Some manufacturers in the past (Nikon and Sony) have chosen to clip a little bit of the blacks, however both manufacturers these days use a bias offset of around 600 now. The key problem with shadows is they get lost in noise.

So when you create a photograph with a digital camera...you have either clipped the highlights, or not. When it comes to preserving highlights, so long as you understand your camera and have a good feel for where it clips, you preserve highlights simply by not clipping them. That does not require ETTL per se...it simply requires paying careful attention to how far the highlights extend in your histogram. (For Canon cameras, adding MagicLantern, and using it's RAW histogram, can GREATLY help in this area.) So long as you don't push your highlights beyond the point where they clip, there is nothing else you need to do to preserve them. They are all preserved. That's it.

The entire reason ETTR exists is because in the highlights, and in the midtones, the signal is strong enough that the SNR is acceptably high. You have low noise because noise is the square root of the signal. The square root of something grows more slowly than that something itself. If your signal is 10,000e-, your noise is 100e-, which also so happens to be a ratio (or SNR) of 100:1. If your signal is 100,000e-, your noise is 316e-, an SNR of 316:1. It's in the shadows where noise can become a problem. A signal of 1000e- has noise of 31e-, which is an SNR of a "mere" 31:1. A signal of 1000e- is actually fairly strong...that is actually probably more like low midtones or brighter shadows in most cases. True shadows can be as little as 50e-, and deep shadows can be as little as 10-15e-! There you have very low SNR.

The problem with shadows is compounded by electronic noise, however. There is noise inherent in every signal, but with a digital camera, you also have added electronic noise. Read noise, as it is usually termed, is compounded with (added to in quadrature) the noise in the signal. If your camera has 30e- RMS read noise, and your signal is 50e-, then your SNR is not 50/SQRT(50), it is 50/SQRT(50 + 30^2)!! That is a minut SNR of 1.62:1! A Canon 5D III has MORE than 30e- read noise at ISO 100. (Which should explain why it's detail in the shadows breaks down and becomes unusable with a mere 3-4 stop push above.)

If on the other hand, you have a mere 3e- RMS read noise, and your signal is 50e-, then your SNR is 50/SQRT(50+3^2), which gives you an SNR of 6.51:1. An SNR of 6.5:1 is actually fairly decent for shadows. (As an astrophotographer, I often work with SNRs smaller than that.) The Sony A7r, and for that matter, the A7 and A7 II, the A6000, the Nikon D800, 810, 600, 610, 750, the 7000 line, the 5000 line since the 5100, the 3000 line since the 3100, etc. all have read noise in the range of about 3e- or less (regardless of ISO). The Samsung NX1 has read noise around 5e-. (There are new sensors being tested that have constant read noise as little as 1.5e- and less at base gain!) If you have any one of these cameras, then careful and meticulous ETTR becomes less of a necessity. ETTR is still recommended, it's just good practice to make the best use of your sensor's dynamic range as you can, but you don't have to be as concerned about pushing it to the utter limits, carefully preserving highlights, while gaining as much signal as possible in the shadows. The shadows will, as you can see from the above, handle SIGNIFICANT pushing in post, and with some effort to restore some contrast to the scene after a heavy shadow push and even a little bit of NR, can largely negate the need to really worry about ETTR beyond generally shifting your exposures to the right...without clipping highlights.

  • I'm still digesting the information. My camera is a Canon 450D. Looking at your shadow noise, I feel like moving away from Canon, but I just got a new lens. Sigh. – dev_nut Jun 11 '15 at 3:42
  • I personally have stuck with Canon because their high ISO performance is very good, and because their higher end lenses are phenomenal. I own the 600mm f/4 L II, and it is a beautiful piece of engineering and glass. I use that for birds and wildlife. That said...for most of my other photography, I've been moving to other brands. Sony has some compelling options, with IQ like you see above, for quite cheap. The A6000 and A7r II are on my list. I may get a Samsung NX1 for shorter focal length wildlife (w/ forthcoming 300mm). A QSI CCD camera for astro is on the way soon. – jrista Jun 11 '15 at 4:59
  • I used to be a pretty loyal Canon fan, but after watching and waiting for years for them to improve their sensor technology, I have come to the conclusion that they simply can't, because they either don't know how, or simply don't care. IQ improvements are being made in leaps and bounds by all the competition, and the best Canon has come up with since the 5D II hit the streets is the 5Ds...which MAYBE gets about half a stop improvement in noise performance. If you want better IQ...don't let brand loyaltee get in your way. Cameras come and go...and with mirrorless, lenses... – jrista Jun 11 '15 at 5:00
  • ...can be easily adapted. Both the Sony A series cameras (E- and FE-mount) and the Samsung NX class cameras, are mirrorless. That means short flange focal distance, which means just about any lens, of any brand, going back many decades, can be used on those cameras with simple and relatively cheap adapters. That opens up a whole different world of image quality that simply isn't available in the Canon line at the moment. – jrista Jun 11 '15 at 5:02
  • 1
    Nikon shadows will be pretty much the same as Sony shadows. They both use the same Sony Exmor sensors. Nikon actually does a better job with their RAW data, as it is true raw, not lossy compressed...so you might be able to eek more performance out of each file. Overall, though, the differences between Sony and Nikon cameras that use Exmor sensors is minimal. Your fine going with either brand. Almost everything out there these days uses Sony Exmor sensors. The exceptions are Canon, which use Canon sensors, and Samsung, which use Samsung sensors. Even MFD use Sony now. – jrista Jun 11 '15 at 22:49
5

ETTR is advocated so that you expose enough until you don't clip highlights

That would be wrong without adding the word "important" or something similar, like in "don't clip important highlights".

so that you can get less signal to noise ratio in the shadows

Not less, but better (actually, higher; as noise is lower).

ETTL is advocated because the shadows preserve better details than highlights

I think "ETTL" is not mentioned in the article at all. The term ETTL becomes outright redundant if we agree important highlights, not just any highlights, should not be clipped.

There is no ETTR or ETTL per se. There is technically optimized exposure, and creative exposure (a whole different story). There is exposure for the scene, and exposure for the main subject in the scene. The following is mostly about the "noise and resolution optimized" approach to "exposure for the scene" for relatively wide dynamic range scenes which we want to capture as accurate and as full as possible in a single exposure.

"Expose to the right" rule is in fact exposing for the shadows without blowing out important highlights (that is, preserving those highlights, like in ETTL :) ) where one needs to preserve a hint of texture; or, in other words, give the shot as much light as possible (through aperture control and shutter speed, mostly; as ISO setting is not a part of the exposure, raising it does not add light, and it is raised when opening the aperture wider and setting slower shutter speed is not an option anymore) without damaging the important highlights. What are those important highlights depends on the scene and is determined by the photographer.

To determine what are those highlights I can afford to blow out I ask myself a question - will the photo be ruined if I let go this or that texture? Next, can I frame closer, or crop, to exclude the areas that will be blown out because the dynamic range of the scene as I framed it is wider than what my camera is capable of reproducing? Or maybe I need to change my position with respect to Sun, or shoot under different light, or to use some fill light (flash, reflectors, etc.), or shoot at a different time; or to use a polarizing or gradient filter, or some combination of the above.

To determine the dynamic range of the scene I use spotmeter to meter from the important highlights and important shadows and see what is the difference between those 2 measurements. If it is more than 8 stops I start to decide if I want to shoot the scene as it is: there is a good chance that shadows will be of lower quality, with more noise, less resolution, and more colour skews than I want them. Yes, some would say modern cameras are capable of much more than 8 stops; but the spotmeter (even my hand-held 2-degree Sekonic) integrate over a certain area, and the in-camera ones are prone to flare on top of that. So, from experimentation I decided that 8 stops is a good number, and will never go for more than 10 stops in a single exposure even with the best current cameras and prime lenses.

It is worth mentioning that histogram, and especially in-camera histogram, is not the best way to determine the proper exposure for the scene that includes light sources and specular highlights because it does not show where the blown-out highlights are in the scene. Most of the cases those specular highlights and light sources need to be let go. Histogram in this case indicates overexposure, but in fact it is not the case. "Blinkies" show the locations of the blown-out highlights, highlighting what areas are clipped in the scene.

If you are shooting raw, one of the important things is to know how much headroom there is between the metered exposure and clipping point in raw. This number varies between camera models and makes, and even ISO settings, from 2.5 stops (equivalent to 18% grey) to 4 stops - and even more for "fake ISO" settings (that is, when the raw data is to be multiplied during the raw conversion). Searching for "exposure meter calibration" should result in several useful hits.

If the scene is 8 stops according to spotmeter or less, it can be exposed based on the measurements from shadow area that needs to be rendered in detail; with exposure compensation that depends on the calibration of your camera (something about -4EV). This allows for close to least possible noise. Same strategy if the scene measures wider but the highlights are not important. If the highlights are important, the measurement is taken from the highlights where you want to preserve some details and texture and about +3 EV is added.

Otherwise, the old school advice still stands. The exposure is based on the most important midtones (something like measure the forehead of the model or the back side of your own hand and add +1 EV; or measuring incident light and adjusting, taking into account camera calibration). Those midtones need to be exposed as high as possible, once again, without clipping important shadows and important highlights. This way the most important midtones will have the best definition, resolution, acutance, colour fidelity, least noise, and will need least editing while postprocessing the image.

  • 1
    This answer goes into detail. It also makes me feel as if I need to learn more to make sense of it all :). It would be great though for future reference to mention which part of your explanation refers to ETTL. – dev_nut Jun 10 '15 at 16:21
  • @dev_nut : added a little more to the text. – Iliah Borg Jun 10 '15 at 18:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.