CAMERA



Working with the S-Log color space.

The basics of log curves and gamma (part 1 of 4).

The development of the gamma curve to HDTV alone would be worth a multi-page article. But we’ll keep it short here: The gamma curve in the camera determines how different brightness levels are transmitted from the sensor to the video image. A sensor has a different characteristic than a monitor. Human vision is also subject to a certain characteristic, which can be represented as a curve. In addition, the brightness perceived by humans increases steeper in dark areas and less steeply in bright areas (curve [gamma] exponent approx. 0.3 to 0.5). In order to bring these worlds together, gamma curves ensure that we can see subjects in the correct contrast according to our sensation on the monitor.



Essentially, current HD and 4K monitors also imitate a tube TV with "analogue" gamma 2.2, since the cameras - initially analogue, later digital - were originally developed for this analogue tube TV reproduction. At the latest with the digitalisation of the video cameras, the developers had the opportunity to implement various characteristics for the recording. Since the cameras could now accept much more tube contrast than the initial approx. 5 f-stops, the question arose what could be done with the "excess" information, especially in the highlights. Or let's summarise it differently: Today's cameras easily "see" 12 to 14 f-stops contrast. Nevertheless, today's television standard (short: Rec709 ) can only reproduce about 6 f-stops with high contrast.

What does it specifically mean? As long as the television standard does not change, we can't record larger contrasts and reproduce them correctly for the human eye. But the cameras can already do much more, so what? Right, but only the cameras can. By bending the gamma curve inside the camera, I can achieve more than 6 f-stops of subject contrast if the sensor delivers it. Viewing on a Rec709 monitor is always a bad compromise, even if many like the flat look. The picture is then displayed with much less contrast than it originally had. One would also have to bend the gamma curve in such a way that the grading would also deliver more.

This leads us to the logarithmic recording options:

LOG:  EXAMPLE S-LOG

Unfortunately, digital sensors work differently than the human eye does. The digital measuring at the sensor is linear (e.g. 8-bit, 256 numerical values). However, the problem is that the human only perceives a doubling of the light quantity as twice as bright - which can also be described with an f-stop (a "light value"). This "doubling behaviour" is logarithmic. If the sensor reaches its light quantity limit (i.e. the photocell is filled and saturated with light), it digitally generates "its" full bit number, for example at 8-bit: "255". From there it goes downwards:


The dilemma is obvious: "Above", i.e. in the bright areas of the sensor, one f-stop is recorded in great detail. The darker it gets on the sensor, the fewer values (binary logic of the conversion) are available for one single f-stop. Even at five f-stops below maximum white, an 8-bit sensor linearly offers only 7 grayscales ("8-15") - what might a subject look like in that case? Another difficulty is that we humans perceive brightness exactly the other way round: We can differentiate well in the shadows, but we find it much more difficult to recognise details in lights - a result of evolution.

We're still talking about an 8-bit sensor, not an 8-bit video recording. Today's sensors have at least 12, 14 or even 16 bit conversion. An example: The Sony F5/F55 or also the FS7 / FS5 have a 16-bit sensor. So the list would look like this, the maximum value that can be generated at 16-bit is "65.535":



The dilemma is obvious: "Above", i.e. in the bright areas of the sensor, one f-stop is recorded in great detail. The darker it gets on the sensor, the fewer values (binary logic of the conversion) are available for one single f-stop. Even at five f-stops below maximum white, an 8-bit sensor linearly offers only 7 grayscales ("8-15") - what might a subject look like in that case? Another difficulty is that we humans perceive brightness exactly the other way round: We can differentiate well in the shadows, but we find it much more difficult to recognise details in lights - a result of evolution.

To honor of the above example, we're still talking about an 8-bit sensor, not an 8-bit video recording. Today's sensors have at least 12, 14 or even 16 bit conversion. An example: The Sony F5/F55 or also the FS7 / FS5 have a 16-bit sensor. So the list would look like this, the maximum value that can be generated at 16-bit is "65.535":


Again: The brighter, the better. It becomes recognisable where the limits of the conversion are. Only a handful of numerical values for an f-stop (here: -14) cannot provide a satisfying image. However, there are still some details visible. In addition, these areas are already deep in the noise of the sensor, but more about this in later parts. Ready? Basically yes. Now we only need to expose the 16-bit as bright as possible, take pictures and can then move on to the grading.

IT WOULD BE SO EASY - WITH RAW

In fact, the F5 and F55 use the system's own RAW recorders to record 16 bits each for R, G and B from the sensor. However, neither FS7 nor A7S or the above mentioned F5 and F55 can record 16-bit via video compression. The most common video compressions use 8 or 10 bit. So throw away half of the dynamic range or leave them with bad numerical values - see above? The trick is : Log. Once the 16-bit values are internally available, the camera can do anything with them, for example bend the values or "take out" only 6 f-stops (the highest = lowest noise) and thus feed the recording of an ENG camera. Why not?

But many users don't like hard-burning highlights. That's why manufacturers like Sony don't start the internal "ENG"-typical Rec709 evaluation right at the top shortly before the sensor's total saturation level, but a few f-stops lower. So ISO2000 of an FS7 becomes ISO800 in Rec709.

With automatic knee or the flattening parabolic film gamma curves (for Sony: "Hypergamma" or "Cinegamma") typical video burnout can be avoided or concealed. This also looks good on a Rec709 monitor. "Log" respectively S-Log ("Sony"-Log) goes much further. S-Log completely leaves behind the characteristics that would be necessary for a video or PC monitor. Following the slogan: We do what we want. The main thing is to preserve the full 14 f-stops. This is also where the biggest misunderstanding lies: log recordings are not a LOOK, but a technical procedure for post-processing. Unfortunately, word hasn't gotten around everywhere yet.

WRONG  LOG?

At the latest now the question arises whether a log format really logarithmically records or only uses the chic and trendy name. Already the Sony FS100, which neither offered a real log recording nor could generate a real log image, came up with picture profiles. These profile settings had misleading names such as "G-Log" or "K-Log" and could record more sensor information. But they did not work logarithmically and were internally limited to 8-bit. Even today, it would be worth considering how seriously different manufacturers actually take the log issue. The idea behind Log is to provide as many grayscales as possible for each f-stop and to record this with a standard video compression of 10-bit (Apple ProRes , XAVC , DnX-HD etc.). How about the following idea:



So you can roughly distinguish the f-stops of S-Log3 in: "Maximum" (> 70 values per f-stop): F-stop MAX to -8; "Average" (> 30):
f-stop -9 to -11; "Nice-to-have"(< 30): f-stop -12 to -14.

In the next part we will compare the still popular S-Log2 with S-Log3, clarify if S-Log3 is noisy and how to expose best.

Related Articles: