The thing is, you have a limited number of bits per pixel, and human perception of changes in brightness fall off logarithmically as things get brighter. So, to make sure you get the most “bang for your bits”, you want smaller quantisation steps for low intensity and bigger quantisation steps at higher intensities. L*a*b achieves this by using a (non-linear) gamma encoding.
In the case of linear RBG, if you use floating point numbers, they consist of a mantissa and an exponent (like in scientific notation), so as things get brighter, the quantisation steps naturally get larger due to the fixed precision of the mantissa. So, using floating point numbers for linear RBG representations is actually pretty important. On Sat, 1 Feb 2020 at 2:03 am, David Vincent-Jones <david...@gmail.com> wrote: > darktable's use of floating point calculation, as I understand it, was > important for maintaining accuracy through L*a*b processing. Is this > still valid as the emphasis in processing shifts to RGB? The reason that > I ask this is to question whether RGB modules still are greatly helped > by the use of OpenCL. > > > ____________________________________________________________________________ > darktable user mailing list > to unsubscribe send a mail to > darktable-user+unsubscr...@lists.darktable.org > > ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to darktable-user+unsubscr...@lists.darktable.org
