The Sony XDCAM EX1 camera provides 8 different gamma curves, each of which can be modified by the user. Unfortunately Sony haven’t published the curves in quantitative form and user opinions of how each affects recorded images is highly variable. Much uncertainty can be removed by quantifying the default characteristics of each of the gamma curves so they can be compared.
The methodology is simple, repeatable, and I think useful. All necessary data is displayed on the camera LCD, namely image brightness (%) and f/stop. The measurement set up requires a uniform constant light source large enough to fill the FOV, with the camera locked down. I used the lens wide to ensure max effective aperture, but the f/stops are relative anyway. Starting with the f/stop that gives 109% I closed down in half stops recording brightness at each change in f/stop reading. The change in f/stop reading is finely repeatable and I took that point (closing) as being the actual aperture value. The process is equivalent to decreasing the test subject's brightness in half stop increments, and because it is all relative it is unimportant whether the f/number is the true value, provided the increments are accurate (which they might not be towards the high f/stop end). I used ND filters to work aperture over a limited range. Each gamma was set to its default setting, gain to 0dB, daylight WB.
In plotting the results the geometric value of the f/number was used in place of the rounded figure which is displayed on the LCD. Any small irregularities in the plotted curves are related to experimental error.
Using this method you can quantitatively determine the effect of changes to the default settings, such as knee (on the standard gamma curves), black gamma, etc.
The default curves for Cinegamma (figs 1 & 3) and Standard curves (figs 2 & 4) are here plotted in linear and log form.
The final figure (5) shows the type of influence that gamma level setting has on the cinegamma #4 curve. Setting gamma level to +20 keeps maximum image level to 95, no matter how bright the subject, and thus absolutely flat highlights (no over exposure and no differentation)
Figure 1: Relation between subject brightness (relative) and resulting image brightness for the four cinegamma curves (compared to Standard Gamma #1): linear plot.
Figure 2: Relation between subject brightness (relative) and resulting image brightness for the four standard gamma curves; linear plot
Figure 3: The data of figure 1 replotted showing image "level" against log of relative subject brightness.
Figure 4: The data of figure 2 replotted showing image "level" against log of relative subject brightness.
Figure 5: The effect of changing the gamma level setting of the Cine4 curve.
Appendix:
Using images to show the effect of a changing a parameter is limited
because one image is just one example. A set of images taken to show the
effects of "twiddling the knobs" are most meaningful to the person who
took them, which is why anyone who shoots video (and film) for a living
always do their own tests. Yes, they start out with guidance from
others, but rather than using that guidance as a recipe they run tests
for the "look" they want for their project. Really understanding how an
image interprets a scene requires knowledge of the original scene, which
we can't get unless we were present. Describing the scene is inadequate
(requires the old "1000 words"), which is why technical tests use
standardised cards. People have tried to communicate the effects of
different gamma curves by subjective terms such as "the brighter gamma",
which really tell us very little.
Graphs show us how the camera translates subject brightness into image
brightness, and importantly how camera adjustments affect that process.
Alister used a graph to explain "knee", much more clearly than using
just words and a photo.
The camera can record images of relative brightness that range from 0 to
109% -- black to "super" white --- I think everyone understands that.
But how much light (subject brightness) is needed to get a specific
image level depends on many things (aperture, filters, image processing
algorithms). How the image is processed is influenced by our set picture
profile, especially the selected gamma curve. In addition, the light
coming into the camera is divided into RGB streams, each processed
separately; best put that aside for the moment and think just of the
total (white) light stream.
So, just as image brightness is given in relative terms ( 0 to 109%) so
it is easiest to think of subject brightness in relative terms. I chose
to make brightness (think: white card) relative to that which gave 109%
when using the cinegamma curves. The curves are very flat at that point,
and I might have chosen instead to plot subject brightness relative to
that which gave 109% for a standard gamma (so then the curves would have
shown subject relative brightness of 5.55 for 109% using cine 4). Or I
could have chosen any other subject level --- the important thing is
that all the curves are relative to the chosen "standard".
Looking at the curves you immediately see the way cinegamma handles
highlight regions. Each halving of relative brightness (1.0 to 0.5, 0.5
to 0.25, etc) is a stop reduction in brightness (on our blank card), so
we see that cine4 results in brighter mid-tone images than cine1 (for
same settings otherwise). At the same time we see that STD1 will give a
more fully modulated image (near saturated) at the same settings, and
because the STD curve is steeper the image contrast will be greater;
bigger change in image brightness for a given change in subject
brightness.
Looking at these graphs we know, when we start shooting tests for our
next project, that if we want to achieve contrast similar to the STD
curves but need to use a cine curve because our subject has a wide
brightness range (e.g. outdoors, white clouds, sunlit and shadow), then
we will have to grade the image in post. If our subject is quite flat
(small brightness range) then a STD curve might be preferable. In
general, the cine gammas capture the most subject data (that is, within
black level and white clipping) but require grading in post because
otherwise the images look flat. If you want camera output direct to TV,
then the STD gammas will be most pleasing (but may have white clipping
and/or choked darks).
Going to the final graph, it shows that adjusting gamma level has a big
influence on rendering highlights, but not much in mid-tones. And in
other curves we see that log curves are more revealing of response at
the darker (relative) range of the subject. Of course this set of curves
on my blog is a limited one and there are many more parameters that can
be plotted (black gamma, for example). And we haven't considered
separately the RGB streams, where (for example) we might be concerned
with how highlights are handled by each of them, or how skin tones can
be distorted at higher zebra (image brightness) levels.
You get a lot of information from response curves and they make it a lot
easier to understand what the available controls do. Of course someone
has to generate the curves and in the same way that Kodak and Fuji
always publish data for their film emulsions (response curves,
modulation transfer function, etc) I think it a pity that video camera
manufacturers don't do similarly.
