QAD HDR: Expose to the Left and Right

Back near the turn of the century, I was one of the early proponents of ETTR (expose to the right) for raw shooters. The classic definition of this is that you place the brightest highlight you want to preserve right near the saturation point of the sensor (255 for any given photosite in 8-bit terms). 

Any highlight recorded at 255 is suspect, as you don't know if that was the proper value or if you just saturated the sensor and bit count for that position. Nikon cameras (lately) start blinking the highlights at about 248, if you want a reference (but remember, that's based upon the JPEG settings).

After taking an ETTR shot, you use your post processing software to "bring up the shadows." There are lots of ways you can do that (black point, curves, shadows slider, exposure slider while bringing down the highlights slider, etc.). Over the years sensors have gotten better and better at still having "usable" data well down in the shadows. Indeed, you'll see some Web sites regularly showing what it looks like to move an exposure slider in a Adobe converter up five or six stops.

The implication is that this is fine. Just don't blow the highlights, then adjust the shadows. 

In reality, it's not. On the best modern sensors, yes, you might get "usable" shadow detail by pulling up the shadows, but it's not at all optimal to do so. I repeat, it's not at all optimal. 

Why? 

Two things: randomness of photons and number of data bits being used. Both work against you getting great shadow detail when you just pull the sliders (or commit slidercide, as Kelby is wont to say). The more noise and rounding errors there are in your deep shadows, the rougher and muddier they'll look when you try to bring them into visibility.

In film, Galen taught me to "bring up the shadows" by using fill light. It's one of the reasons why I always traveled with a flash and a reflector. Likewise, Galen also advocated pulling down the highlights for film using graduated NDs. 

In digital, the dynamic range expansion rage became HDR. Take a long series of shots and then let your computer's CPU get a workout trying to make sense of all the overlapping data. I've noticed that Kelby, for instance, has backed off to three-shot HDR sequences (-2, 0, +2EV). That's understandable, because shooting something like -2, -1, 0, +1, +2 provides you 66% more data but likely no additional observable benefit. You're just going to tax your CPU/GPU to make sense of all that additional mid-range data stack.

But we're still talking about a maximum of a four stop gain, and probably far less than that if the 0 exposure is from the meter (it's rare that you have equal sized needs in the highlights and shadows from the metered value). 

What's the proper choice? Expose to the Right, then Expose to the Left. 

In your first exposure, you want the highlights preserved (ETTR). Many of us took to using something called UniWB to get more accurate in-camera histograms for raw files so we could insure that we preserved highlights in all three channels (red rock landscapes at sunrise/sunset are notorious for blowing out the red channel but not really showing that accurately on the default in-camera histograms based on the embedded JPEG preview image). But just zero out the JPEG stuff (Picture Control should be Neutral, no extra contrast, clarity, sharpening) and get your white balance right and you should be close.

  1. However you do it, make an exposure that preserves all highlight data. 
  2. Then make a second exposure that preserves and promotes the shadow data.

In your second exposure, you want the shadows preserved (ETTL). This is more complicated than ETTR, because it's going to take some advanced judgment skills on your part. It's okay for black to be black (0). But the question is what level of shadow detail do you want to render well, and how do you interpret that word "well"? 

Personally, any line up the left edge of the histogram is something I try to avoid with any ETTL exposure. But exactly where the values on the left side of the exposure belong is a total matter of personal assessment based upon what you're trying to do in post. The higher up the histogram the shadows are, the less noise they have (;~). But that may also start putting shutter speeds into marginal levels for you if you're not shooting static landscapes with a tripod, and it might cause issues with the HDR software you use to combine images if you push that exposure too far. Which leads to: 

We may need a third exposure. I find that to be true if the exposure value changed by more than 3 or 4 stops between my ETTL and ETTR exposures. Why? Because it gets more and more difficult to properly evaluate multiple exposures and blend them the further they are apart. A 2 or 3 stop difference actually seems to work effectively in Adobe converters and other software. More than that, not so much; a third exposure will help the blend. That's even true if I do manual blending (e.g. combine exposures on horizon lines). 

So:

3. If #2 was more than 3 stops off from #1, consider shooting another exposure in the middle.

You might note that many cameras, for instance the Nikon DSLRs, take only two shots when they create an HDR JPEG. That's because they're basically using something similar to the technique I describe here (and punting on getting more than a few stops of usable gain). 

text and images © 2018 Thom Hogan
portions Copyright 1999-2017 Thom Hogan-- All Rights Reserved
Follow us on Twitter: @bythom, hashtags #bythom, #dslrbodies