Sony recently asked Adam Wilt and I to speak about the new FS700 camera at CineGear. My portion of the presentation examined how to profile dynamic range and determine how the various gamma curves changed the camera's response. Read on to see what I discovered...
I've worked as a consultant, both paid and unpaid, to DSC Labs, manufacturers of the best HD color charts in the world. This has some benefits, such as having access to charts that aren't as easy to find as they should be. One of these charts--the one I consider indispensable when learning how a new camera functions--is the DSC Labs Xyla 20-stop dynamic range chart:
The theory behind this chart is simple: each of the 21 steps is exactly one stop denser than the one before it, which makes it dead easy to look at the image of the chart on a monitor and see, instantly, how much dynamic range a camera has with its current settings. And--even better--looking at the pattern on a waveform monitor will show exactly how the gamma curve is shaped, revealing which portions of the curve are compressed and which are expanded.
And, as every modern camera has the equivalent of at least a half dozen film stocks built into it, this chart makes it very fast and easy to see how the contrast of each of those options varies.
There is no faster way to learn how a gamma curve allocates dynamic range.
The eagle-eyed among you may have noticed that I said the chart covers 20 stops of dynamic range and yet it has 21 steps. That's because we clip the first step and count stops from there. (You'll notice that the bottom of the chart pattern is labeled 1-21 in steps, but the top of the chart is marked in stops and is marked 0-20.) That's because the only consistent exposure reference point an HD camera offers us is the point where the photosites on the sensor saturate, or clip. In film we have a number of density reference points (D-min, D-max, middle gray) but those don't apply to digital sensors. The one thing we know about every sensor ever made is that, at some point, you'll max out the exposure and the sensor will give up and show featureless white.
From my presentation notes:
"The only reference point a sensor offers us is where it saturates, or clips. Film has solid reference points: maximum density, minimum density, and 18% gray. In digital cameras 18% gray, or middle gray is an arbitrary value based on the particular gamma curve used. We can't use middle gray as a reference and measure latitude above and below; we can only reliably measure how many stops down from clip a particular tone occurs. Middle gray moves around depending on the gamma curve used."
What this means is that one gamma curve may plot 18% gray at the 3rd step below clip, while another may plot it at the 5th step below clip. Both will make a properly exposed 18% gray card look middle gray, but the latter will offer more overexposure headroom--along with an increase in noise.
Noise is the single biggest determinant of how fast a camera can be rated. The less noisy it is the higher the ISO, and the more overexposure latitude can be made available because 18% gray can be pushed closer to the noise floor. This allows the gamma curve to allocate more steps to highlight detail.
I'm quite proud of my own contribution to this chart's design: contrast is so high between the brightest step and the darkest that flare can dramatically affect shadow results, so the wedge shape minimizes the size of the brightest steps to reduce flare that may artificially boost the darkest steps. (Also, the chart comes with sliders that can block the brightest steps and make the darkest steps easier to view. That part wasn't my idea.)
In theory this chart should be used in a very, very large, completely black room--but modern cameras can't see the full range of this chart without using HDR, so we were able to shoot these tests in a dark stage (at Meets the Eye) and get perfectly usable results.
Different manufacturers choose different values for 18%, or middle gray. There is some math involved but I've also been told it's a matter of taste. Most say middle gray is in a range between 38% and 42% on a waveform monitor, so I've chosen 40% as a happy medium.
The Rec 709 spec that all HD devices are based on was designed to hold only about six stops of dynamic range. Today's cameras capture a much greater range of brightness, typically at least ten stops from clip to black. Highlights and shadows can be compressed because that's what our eyes do anyway: we are most sensitive to contrast in mid-tones, the rest can be cheated. As long as mid-tone contrast is correct, contrast in highlights and shadows can be reduced--which allows the camera to cram more stops into a smaller space.
Ideally there should be more space between the middle steps than in the steps at either end. I've arbitrarily selected a range between 20 units and 80 units for this example; most gamma curves compress contrast above or below these points.
This isn't a good example of highlight compression but it is a good example of shadow compression. Notice how the steps flatten out as they head toward black. Some of this is simply how the sensor responds to light, but some of this may be the work of the gamma curve. Whereas the original Rec/ITU 709 spec may have allowed for only about 3-3.5 stops in the range between 18% gray and black, we can see 6 or 7 here.
The idea is this: modern cameras capture at least ten stops of dynamic range, but they are forced to fit those into a range originally designed for only six stops because that's what the Rec/ITU 709 standard dictates. Changing this standard to accommodate the increased dynamic range of modern cameras means scrapping every device manufactured in the last 12 years, so it's generally more economical to work within the standard than to create a new one.
This means modern cameras must play games to make their images look their best within these limits, so they mush the brightest and darkest stops together to cram 10+ stops into a 6 stop range. The most important area of the curve, where our eyes see the most contrast, is in the middle range. If those steps aren't nicely spaced then the image will look flat.
From my presentation notes:
The first step is just clipped, and we count down until we hit 18% gray at 40 units. The first step is zero, or clipped; the second step is one stop from clip; the third step is two stops from clip, and the fourth step is three stops from clip. That's where we hit 40 units, so we can say that this gamma curve offers us three stops of overexposure latitude above 18% gray. I see seven additional stops below 18% gray before the darkest visible steps dissolve into noise.
The last step visible on the waveform may not be the darkest step viewable on a monitor. This depends on a number of factors, such as the quality of the monitor and whether the viewing conditions are suitable. A lot of ambient light may make the darkest steps harder to see.
Here I've boosted the blacks to 20 units just to see what's happening in the deep shadows. I count 14 stops from clip to the last visible step. Those last few stops are most likely not usable and shouldn't be counted on.
I divide these stops between "paycheck" stops and "gravy" stops: paycheck stops are the values that need to be there or your paycheck stops. Gravy stops are nice if they are there, but if they aren't you're still going to keep your job. The last few stops we saw on the previous gamma curve are, for me, gravy.
The Xyla chart is a great tool for showing overall gamma response, but we should always look at the image on the monitor to see what's realistically visible.
On the next page I lay out all the FS700's gamma curves, in all their glory...