The highlight clipping thing

March 28, 2017  •  Leave a Comment

There has always been a debate about how to expose correctly.  Back in the day, as they say, when the choice for most photographers was either to shoot colour negative or colour slides, the common wisdom was to under expose to preserve highlights in slide film and expose normally for negative films.  Slide films could render only a truncated exposure range around 6-7 EV while negative films could go 12-13 EV.  The ability to render detail and colour in the brightest and darkest portions of the image was referred to as the exposure latitude, a term I like.  Many books have been written about how to expose for film using spot or matrix metering, zone systems and what-not.  It was important to be able to pick out a middle grey tone in a colour scene, like grass on a cloudy day to help the photographer compare what the camera metering system was telling him vs what he wanted to achieve in the final image.  Finally, film has another interesting feature:  the abrupt changes into highlights was mitigated into a smooth transition in what is called the shoulder of a tone curve.

Then along came digital sensors and the exposure debate escalated.  Digital photography changed a photographers thinking about exposure in several important ways.  First of all, the digital world comprises of zeros and ones, none of this tapering tone shoulder shenanigans we were used to with film.  This meant that if you did not expose for highlights properly they abruptly turned an ugly white.  A great example of this is facial highlights.  Nobody likes to see white blotches on faces.  A second change was the feedback as the photographer could review the photo immediately in a little LCD window on the back of the camera and see the exposure.  Chimping became a reflex.  On most cameras the blown highlights could be seen as "blinkies" that would flash on the camera LCD.  Finally, the photographer could refer to a RGB histogram, which would show the tonal distribution for each colour channel.  With all this feedback you would think exposures would be pretty straight forward and the debate would recede into a background hum.  But then along came ETTR.

ETTR or Expose To The Right was first promulgated by Michael Reichmann in 2003 to address the combined issue of digital linear capture and electronic signal to noise.  Human vision and the film response to light is logarithmic.  We see each doubling of brightness, or the number of photons, as an even increase in brightness.  This is not how digital sensors work.  They basically count photons.  If we see a scene and divide it into 6 even tones we see 1, 2, 3, 4, 5 and 6.  A camera sensor sees 1, 2, 4, 8, 16 and 32.  This probably seems familiar because this is how apertures, shutter speeds and ISO work.  What should be evident from the 6 tone example is that the camera sensor can differentiate 16 different subtones in the brightest part of the image but only 1 tone in the darkest part of the image.  Without diving into details about electronic signal to noise the take-away is that the noise will be much more evident in the darkest tones, where you will only see very dark, with white contrasty speckles (or red, green and blue noise speckles).  In order to see more tonal detail in dark shadows and to minimize noise ETTR suggests that you should increase the exposure so there are more photons hitting the sensor from the shadows.  This makes total sense as long as you make sure you do not blow out the highlights, which Michael Reichmann made sure to state.  If the tonal range of the image where the photographer wanted to retain colour and detail exceeded the dynamic range of the sensor then obviously ETTR would not apply.  This can be seen in the following two histograms, the first from a typical high contrast landscape with sunny sky and shadows, and the second, from a lower contrast evening landscape.

As you can see, there is some room to increase the exposure in the second histogram, while in the first, the brightest tones will start to blow out.  Since ETTR was a new concept to many photographers they tended to take the a literal view and forget all about the clipping highlights thing and there were lots of arguments in web forums about ETTR.  It  really is pretty simple:  first decide it you want to hold onto the highlights and then if you want to pull detail from the shadows.  Photos of people often present this dilemma where you need to control the highlights on skin but still want detail.

In the first image the exposure shows detail in the shadows at the expense of the highlights.  The second exposure has tamed the highlights, but now some of the shadow areas have almost gone to black.  In order to get the best out of a high contrast image like this one, we want to feather the exposure high enough to just be able to control the highlights in post processing and bring up the shadows.

But here's the thing.  You can't quite believe what your camera is reporting.  Your sensor is recording somewhere between 12 and 16 bits of information for each pixel, which is then being compressed down to an 8 bit jpeg file.  It is this jpeg file that is being used to generate your histogram, the blinkies and the image on your LCD.  It turns out that the raw files, which still have that extra information, have quite a bit more exposure latitude than the jpeg in the camera.  Also, the camera manufacturer, catering to the lowest common denominator, is being quite conservative with those blinkies and the histogram.  This is illustrated by this basic landscape, exposed at base ISO using matrix metering and a +1/3 EV bias.

histogram 3  

I don't know about you, but this looks pretty scary to me!  The sky has gone white and the blue channel is crowding both sides of the histogram.  This can't be good.  But wait, let's take a look at this in Lightroom with default tonal settings.

Landscape 1

So, what has happened?  It looks like the camera jumped the gun a little here.  The 14 bit file has a lot more information.  Also, the post processing software, lightroom, might be trying to help out a bit.  When one channel, in this case blue, blows out, the rendering algorithms cleverly try to guess what blue, if any, to show, based on information in the other colour channels.  What this all means is that we have more exposure headroom than we thought.  Here is a spectrum of exposures of the same scene, from -1 EV to +2 EV.

EV

It turns out that even the image exposed +2 EV still has quite a bit of information in the sky, but as shown, is a little washed out in the sky.  Here is the +2 EV shot after some processing in lightroom.  

Landscape 2

It turns out that the +1 1/3 EV image still contains all the highlight information to evenly render the sky.  Here is what it looks like after some tonal ajustments in lightroom.

Landscape 3

But wait, I'm afraid there is more.  In a scene like this the camera will usually nail the white balance.  In evening lighting or artificial lighting this is not a given.  Changes in the WB move the red and blue channels in the histogram.  If you are not careful you can blow out either of these channels if the WB is way off.  You can see it in this example, showing how the histogram changes with WB.

Correct WB Too warm Too Cool
histogram 4 histogram 5 histogram 6

If the camera was misrepresenting the WB as too warm then you would be tempted to under-expose to make sure you did not blow out the reds.  If it is too cool, then you might be tempted to over-expose and blow out the reds.  Note how the greens do not move around too much.  Of course, this behavior changes from image to image.


Comments

No comments posted.
Loading...

Archive
January February March April May June (1) July August (1) September October (1) November December (1)
January February March (2) April (1) May June (1) July August (3) September October November (1) December
January February March April May (2) June July August September October (2) November December (1)
January (1) February March April May June July August September October November December
January February March April May June July (1) August (1) September (1) October November December
January February (2) March (1) April May June July August September October (2) November December