[argyllcms] Re: Writing self tests for gamma correction

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Wed, 27 Feb 2013 10:19:54 +1100

Richard Hughes wrote:
> If a display is specified as "gamma 2.2", does that mean that the
> XYZ->Y component follows a 2.2 curve,

Gamma is about the native response, so it's the device->light characteristic.

> or has each RGB channel got it's own gamma which is then averaged?

Each channel has a gamma. What you do to arrive at a summary number
is up to you :-)

Some of the television enthusiasts like plotting log(out)/log(in) vs. input,
since that gives them a graph of "gamma" at each input level. This is not
without precedent, since the term "gamma" (I think) comes from
the film world, and is the slope of the film density (which is a log
quantity) vs. log of the light film exposure.

A display that perfectly corresponds to a power curve will have a straight
horizontal line log/log vs. input plot. I'm not so sure about
such a graphs suitability for measuring or critiquing display
behaviour, since there is no evidence that it is perceptually linear.

> If it's an RGB measurement we
> calculate gamma with, how do we map from XYZ to RGB? Also, what
> white-point would we use? All the "gamma checker" images decompose
> this into sRGB gamma, but I'm not sure I get the logic why.

I'm not sure what you mean. A display channel that has a perfect gamma curve
response will have light output that is governed by the equation

        light level = in ^ gamma.

You could measure light level as Y, but you'll get better sensitivity
for individual channels by using a vector sum of the XYZ of each
channels maximum. Or simply use the Y vs. R=G=B input if you are
after a single number.

> For my uncalibrated DreamColor, Y is pretty much exactly 2.2. For my
> uncalibrated Thinkpad T510, my Y gamma is 2.7. Ouch.

According to Charles Poynton, such numbers are rather sensitive to the
near black behaviour (depending on how you fit the power curve).

The Argyll numbers are simple but are intended to be more perceptually
related, by being computed from out/in at 50%. Comparing gamma
values for displays that aren't actually a power curve, is basically
a pointless exercise due to the multitude of ways the number could
be arrived at.

Graeme Gill.

Other related posts: