Use sRGB transfer function instead of plain gamma (better accuracy for lower brightness)

[softhack007]

It seems that sRGB has a better way for preserving low-brightness colors, compared to "plain" gamma correction.

As the gamma (and inverse gamma) LUT calculation is not in any hot path, we should try if the sRGB pedant holds its promise to provide smoother transitions at low brightness.

https://en.wikipedia.org/wiki/SRGB#Transfer_function_(%22gamma%22)

The math looks easy, by just replacing the "pow()" expression with the sRGB transfer function.
It's important to also adjust the inverse correction to use the inverse sRGB transfer function.

Edit: the sRGB transfer function uses a gamma exponent of 2.4, the resulting curve approximates a gamma=2.2 curve. I thinks it's possible to derive the correct scaling and offset coefficients for other gamma values - the wikipedia article shows the basic equations needed.

[DedeHai]

that is a good suggestion, I ran extensive experiments on this already and found virtually no benefit - I did use various versions based on the CIE curve with varying "linear curve indices" and also tried different approximations. Depending up to which index the linear part goes (I even tried much steeper slopes) there is slightly better "dark end" color preservation but you also loose "gamma" in that section, sometimes distorting colors even more. It also brightens the dark end of the spectrum so "fading trails" look weird.
There is no universal way to preserve dark colors except for just dropping them to black if they can not be displayed - you can check what that looks like using Blaz's dev branch.