[argyllcms] Re: Gamma measurement discrepancy.

Steve Morphet wrote:
I'm a beginner trying to set up an LCD monitor (Samsung 245T) using Argyll and a Spyder 2, on Linux. I'm puzzled by the difference between the gamma value reported after calibration by dispcal -r, and the gamma values measured using various online squinty test patterns. I wonder if anyone can explain the differences that I'm seeing, and perhaps tell me if I'm doing something wrong. 

Well, I've had this discussion a few times, and it comes down
to the question: what do you mean by "gamma" ?

A "gamma" curve is typically thought of as an ideal power curve,
but no real world device has the necessary zero output at zero
input to be able to match such a curve. The result is that
there are countless ways of matching a real world
curve with the ideal one. So your test patterns may be
approximating it one way, while Argyll is approximating it
another.

Argyll's approximate reading is simply the gamma of the ideal
curve that matches the real 50% stimulus relative-to-white output level.
I think this is a reasonable (and simple) approximation, because
it matches the overall impression of brightness for an image.
A more sophisticated approximation that could be adopted would
be to locate the idea power curve that minimizes the total
delta E of some collection of test values (what distribution
of values ? What delta E measure ? How can delta E be computed
if the overall device color response isn't known ?).

There are many other ways in which it could be done, and any
such approximation may have a quite different numerical value,
even though the visual result is very similar. This is because
the numerical power value is very sensitive to what's happening
near zero, the very point that is non-ideal. Consider
the sRGB curve for instance. It's technically composed
of a power curve segment with a power of 2.4, but when
combined with its linear segment near zero, has an overall curve
best approximated by a power curve of gamma 2.2. Matching
the 50% stimulus would result in yet another slightly
different approximation value of about 2.224. All these different gamma
values represent curves that are very visually similar.
(See <http://www.13thmonkey.org/documentation/fonts/srgbgammacalculation.pdf>).
So, my question is, why the big difference in gamma figures, and which ones should I believe? I appreciate that dispcal is only reporting an approximate value, and that there is probably also some error in my reading of the online charts, but surely that can't account for so much difference. And why did the dispcal gamma increase when the calibration was loaded, yet the chart gamma seemed to fall? 

I wouldn't look to good matches in estimated gamma value. dispcal
by default translates a requested gamma (-g) into the equivalent
real world curve with the same 50% stimulus output as the ideal one.
This seems to lead to reasonable visual matching to an idealized curve.
If you want to set the actual power shape of the curve you can
use -G, but be warned that the visual result will be very sensitive
to the devices response at black and the way the black level
value (input and/or output offset) is combined with the power curve
(-f parameter).
The calibration does seem to have improved the way things look on my monitor, but can I trust it? I feel like the proverbial man with two clocks. 

Personally I stick to trusting the maths, and take the "squintly"
visual charts merely as broad sanity checks, since I never know how
they have been created, or what assumptions they are making.

Graeme Gill.

Other related posts: