I have personally found the concept a bit confusing, and was certainly struggling to understand logically how to modify gamma when calibrating, and what difference it made anyway. So I did some reading via various sources (see list below) and have attempted to summarise some of the key points below, particularly on the practical side of what it means to the picture and how it might be addressed during calibration. I make no promises that what I’ve written is accurate – so please do correct me where I’m wrong!!
One point that seems to have been discussed at great length is what gamma should you try to aim for, or, and this is a different question, what gamma should be considered the “reference” that will show you what the producer intended?
I think the conclusion I could draw was that in general, content is mastered on professional CRT monitors with gamma of 2.5. This ought to be the standard. However, because most viewing is done with some ambient light, a compensation function with gamma 0.45 is applied during recording/encoding, and it would appear that to reverse this you need a gamma of 2.22 (being 1/0.45). However, the 0.45 gamma correction is undercompensated deliberately for CRTs to account for the increased ambient light where it is viewed compared to the dark rooms in which the content is typically mastered.
Therefore, if you view in a dark, light controlled room, you may well be better off aiming for display gamma around 2.5 to get closer to the standard. However, if you view in “normal”, lighter conditions, you might prefer gamma nearer 2.22 or even lower in order to compensate for your perceived response being changed by the stray light content entering your eyes.
Furthermore, “gamma” as used above implies average gamma across the range of stimulus, and sort of assumes its constant for total light output, and probably constant for each of Red, Green and Blue as well. This would be ideal, but in my experience isn’t particularly likely without a lot of tweaking and/or luck.
calzone
04-04-2009, 01:17 AM
Simple Gamma graphs
The eye’s response to light stimulus isn’t linear, it follows a power law. In other words, it has an equal response change to equal ratios of brightness – if light stimulus goes up by a factor of 10, the eye would perceive it as, for example, a doubling in brightness. This would be true if it went from 1 unit to 10 units, or from 10 units to 100 units. This means that the eye can be very sensitive to small absolute changes in brightness when things are dark (because these are actually large ratio changes), but is pretty insensitive when things are bright (because the ratios are small despite the larger absolute changes). I imagine this is pretty useful in nature, because it means we can see well at night but don’t get horribly blinded at noon. I would say noon day light in the UK is maybe only twice as bright as the light at dusk to my perception, but in fact the real brightness is probably nearer 100x brighter – that’s a power law response in my eye’s perception working.
Turns out, probably fortuitously, that the original TV technology, CRT, emits light in response to stimulus in a power law fashion, much like the eye perceives it. This is quite handy…. Apparently CRT’s have an average gamma of around 2.5, and this is why studios continue to master to this standard in light controlled rooms, I assume.
Digital TVs tend to have different, and controllable gammas, because the way the light is generated is different. So its actually worth working out how to tweak gamma, because it a) isn’t likely to be automatically near a standard level and b) it does make a difference to how you see the picture.
If pictures are encoded with a gamma compensation function of 0.45, and then displayed through a display with Display Gamma of 2.22, then what comes out is a power function with the two gammas multiplied i.e. 0.45 x 2.22 = 1.0, which is a straight line.
In other words the TV then emits brightness more or less as it went into the camera in the first place (the blue line).
calzone
04-04-2009, 01:26 AM
If Display Gamma is different, the TV output will be a curve either above or below the straight line, see below (straight line is now the green one…).
<see picture>
The higher the gamma the darker the image on average, but also the more the image increases in brightness per unit increase in stimulus (i.e. the slope) in the mid- to high- range, in other words it will appear to have more contrast in the midtones, but blacks will be less distinguishable and may appear “crushed”.
This might help explain why there are people who like to have a relatively low gamma at low IRE (below 20IRE) and a higher gamma higher up – get good black details but keep good contrast a the higher end.
What you probably want to avoid is very high gamma at the low IRE and low gamma at the high IRE – black crush and white crush all in one….
Looking at the low end of the stimulus scale:
<see picture>
The difference at 5 IRE between gamma 2.5 and gamma 2.10 is that brightness increase from 3.4% to 5.9%, i.e. 70% higher. In a fairly light room, I wouldn’t be surprised to find that the eye struggles to distinguish shadow details on screen at the high gamma, but the 70% extra brightness would really help. In a light controlled and darkened room, the extra tuned sensitivity of the eye can probably handle the smaller changes in brightness at higher Gamma.
calzone
04-04-2009, 01:28 AM
At the higher IRE end of the scale the plots look as follows:
<see picture>
The higher gamma response has a much steeper slope, and so brightness increases faster for each increase in stimulus than it does for lower gamma – this appears as a more contrasty picture.
Roughly, the flip in greater contrast occurs somewhere around IRE 40:
<see picture>
I think this is useful to know – you should be able to see changes in your picture in real viewing based on these effects, and you can choose a gamma target or profile to fit your viewing conditions etc.
I think that while you generally want gamma to be relatively constant (though I can see why you might want low gamma at low IRE and high gamma at high IRE), you also do want to look at the gammas for R, G and B separately, and hope they remain constant throughout the IRE range too.
For example, one of the ways you can tell if you have white clipping is if your gamma drops off sharply at 90IRE – this would imply that brightness is not increasing very much from 90 to 100IRE, perhaps because your screen/projector has run out of light output capacity. Its quite likely however, that average gamma can stay ok at 90IRE, but one of your colours is actually dropping off in gamma, and you’d only see this by looking at the RGB plots and not just the average gamma plot. As I understand it one reason for this is that Green contributes more to your total light output than the other two, and so if green has a slightly higher gamma at 90IRE, you could easily have a very big drop off in Red or Blue gamma and the average gamma would look fine.
This would be a problem because it would screw up your colours – the amounts of each colour being mixed will differ through the stimulus range if the gammas vary significantly for each primary because they won’t have the brightness the signal expects. As a result, instead of having a colour of constant hue vary only in brightness, you will get brightness variation AND a hue change. I don’t know how significant gamma variations need to be before this becomes very noticeable, but worth working on it during the calibration.
One other observation: if black level is set by the Brightness control, and top white level is set by the Contrast control, then the total gamma from black to white is fixed. That means that if you have a low gamma at low IRE, you must have a high gamma somewhere else and vice versa. Any gamma tweaks at say low IRE will have a knock on effect on gamma at a higher stimulus. This is particularly noticeable when using the Pioneer Kuros ISF 9pt gamma settings in Controlcal – it can have unpredictable effects, and isn’t capable of affecting only the point you select.
calzone
04-04-2009, 01:29 AM
Tweaking Gamma during Calibration
Let’s say you have an ok greyscale and flat gamma at say 2.10 until 80-90IRE, where red and blue gamma drop to 1.9. How could you fix that?
1. Check that you haven’t set Contrast too high, beyond the ability of the TV to pump out the blue or red colours. If the TV isn’t physically capable to stretching that far, the only fix is to reduce Contrast.
2. If Contrast isn’t the problem, you will need to tweak the Red or Blue gains. Generally a lower gamma at high IRE means that the rate of increase of the colour brightness is very low – it is not getting much brighter from 90 to 100IRE. To improve this the solution isn’t necessarily obvious however. You might think that you need to increase the Red Gain to get more red light output at high IRE. But this isn’t really what gamma means – in fact lower average gammas are actually brighter…. Gamma is about the rate of change of brightness. And the Red gain affects red in the range 50IRE to 100IRE in not very predictable ways. It may actually be that what you need to do is reduce the Red Gain, because this may reduce the overall brightness of Red and may do so mainly in the 70-80IRE range, which might actually end up with the difference between 80 and 90IRE brightness being bigger, and thus giving a higher Gamma for Red at 80 and 90IRE. Having said all this, because gamma at say 90IRE depends on what gammas have gone at lower IRE, its is definitely not obvious what the solution might be. You might screw up your greyscale by changing Red gain for example, and in fact a small change in Red Offset could change the low end gammas slightly, leaving more room for brightness increase in the high IRE range, resulting in an increase in red Gamma at high IRE.
3. Use of dedicated gamma controls – not sure how this works exactly but because the displays are digital, think that the set can apply some kind of digital signal processing to change gammas directly. This works, but not without an impact – I struggle to see how any change to gamma at a specific point for a specific colour won’t affect that colours gamma elsewhere and the greyscale. So it will need a lot of iteration and tweaking. Shifting the whole range of colours and IREs reduces this problem a lot.
Note: these are my conclusions, and assumptions from my own reading and playing around. If I ‘ve got it wrong, please correct me!!!
