Originally Posted by rader
Thanks for the tip Bill - it would make things much easier if I figured out a way to calculate gamma without measuring black.
Shawn - Looking at the sRGB specs, and trying to back into what some of the Rec709/SMPTE-C documentation says, seems to indicate to me that a gamma function is supposed to include the non-zero black level as part of the foundation for determining the theoretical gamma values. This was what Jeff was indicating. However, if Poynton is correct, then what you really want is for 99% stim to be JND* from 100%. Using CIE Luv, one could do a rough calculation of this (L*), but given the controversy over Delta E*, I'm not sure that this would necessarily produce a better answer than the analytical shorthand that Jeff (and I) use. It would certainly be more computationally intensive.
Alternatively, one could notice that when you normalize, you are usually using a fairly high number (reference white with my H77 runs between 400 and 500 cd/m^2). At that point, the 0.2 cd/m^2 black level is really a rounding error when we talk about normalization. If you use a curve-fit method (e.g., log-linear regression, preferably with >15 measurement points), then the rounding for black level and normalization pretty much get washed out.
So, basically start from 100% and calculate JND down to 0%, or go with a regression method to avoid having to measure a 0% value (even though it doesn't get used in most people's calculations other than to tweak the normalization process).
By the way, you should link directly to Poynton's site, rather than to a mirror. He's got some great stuff there that folks who are using your stuff ought to read.
* JND = Just Noticeable Difference