As some of you are aware, I firmly believe that the common wisdom of ETTR is bad advice that, when practiced
as commonly described, is far too prone to overexposure and heartache.
Instead, I advocate starting with a traditional “proper” exposure, such as determined by metering
off a gray card. If the histogram shows no blocked shadows or clipped highlights (or blinkies in the preview), leave
well enough alone. Only if the dynamic range of the scene is such that shadows are blocking or the highlights are
clipping is it profitable to deviate from such an exposure. When making adjustments, be aware that shadows get
noisy and lose detail and that highlights suddenly transition to no recoverable detail whatsoever. Either make
whatever sacrifices you feel best fit the scene; alter the contrast using traditional techniques such as altering the
lighting; or use HDR.
The common wisdom behind ETTR is that, due to the linear nature of encoded data, the lower stops carry less
information. By using a brighter exposure and compensating in post-production, one can capture additional
information, leading to more shadow detail and less shadow noise. This effect is carried throughout the dynamic
range, of course, but is most pronounced in the shadows where the least amount of information is recorded.
I do not dispute the theory behind ETTR nor claim that the effect does not exist. What I do dispute
is the real-world usefulness of the technique as well as the degree of enhancement to be gained. Considering that
application of the technique requires one balance right on the cliff of overexposure and that additional post-processing is required, I consider the benefit to be far overwhelmed by the risk and therefore the technique not
worthy of recommendation.
Worse, even the theory can only possibly apply at base ISO, as it is exactly equivalent to lowering the ISO.
Above base ISO, it simply doesn’t even apply. The signal amplification will have a far more detrimental
impact on image quality than any linear digital quantization effects could possibly have. (Of course, pulling detail
from underexposed high-ISO shadows is properly known to be insane, but, if you have the dynamic range for ETTR
at high ISOs, you also have enough light to shoot at a lower ISO.)
One should certainly be aware of the theory so one might apply it in suitable circumstances — such as,
for example, when photographing the proverbial black cat crossing the asphalt road at midnight. But, at other times,
the closest that the theory should get to practical exposure is the observation that shadows tend to get noisy and
lose detail — and equal weight should be given to the observation that clipped highlights are
unrecoverable.
Enough preamble. A theory is useless without tests. To that effect, I performed the following experiment.
I shot a ColorChecker with a 5DII and the 180 L in open shade. I defocused the picture so as to smooth out the
texture of the card itself and thereby make noise easier to distinguish; otherwise, the macro clearly resolves each
little bump and pit in the surface.
I made two exposures. (Many, really, but selected only two for this demonstration.) One was metered off the
middle gray patch of the ColorChecker (which, according to my i1 Pro specrophotometer is XYZ: 19.2 20.1 16.5, D50
Lab: 51.9 -0.95 0.28 on this particular ColorChecker); the other was overexposed by exactly one stop. I used f/8 and
ISO 100 for both shots; the metered exposure was 1/30 second and the overexposed shot 1/15 second.
With the following exceptions, I processed the two pictures identically in Adobe Camera Raw. White balance I
set (to 6850K, -4 tint) from the neutral square under the yellow square. All development sliders were set at the
factory defaults, except shadows were set to 0 (and the other exception noted below). I selected the flat tone curve.
All sharpening and noise reduction sliders were set all the way to the left. I used the “Camera Faithful”
picture style. Everything else was set to factory defaults. Images were opened as objects at their native resolution in
8-bit sRGB space. The full-frame pictures I Bicubic resampled to 700 pixels wide; the 100% crops are as-is. For both
I stretched one canvas before placing the two side-by-side. Finally, I used the “Save For Web” export
at JPEG quality 69.
Lab 52 0 0 = sRGB 124 124 124. The middle gray patch with the metered exposure read roughly 143, 143, 143.
Adjusting the slider exposure to -0.45 resulted in a proper exposure. This is consistent with my experience of how
this particular camera tends to overexpose by roughly a third of a stop, give or take, and is consistent with reports
I’ve heard about other 5DIIs. Therefore, the one frame I processed with -0.45 exposure and the other -1.45
exposure.
I will not claim that there is no difference. I expect most people who examine the following pictures with a critical
eye will probably be able to tell which is which. I will state that, as far as I’m concerned, the
difference is entirely insignificant. Remember, these were processed with no noise reduction whatsoever and tone
curves (etc.) that emphasize shadow noise much more than those normally used in artistic work. Not to mention that the subject itself is as textureless as a photographic scene ever gets. Even in this
worst-case scenario, were I to make 24″ × 36″ prints from these exact files, you’d still
have to stick your nose in the print to even have a hope of maybe being to tell which is which.
First (assuming I don’t screw up the image attachment, as I’m wont to do) is the full-frame shots.
I’ll post the 100% crops of the darkest neutral square immediately after.
Cheers,
b&
, Mar 10, 2010; 07:45 p.m.
