Originally Posted by ConnecTEDDD
A 17-Point Cube Profiling (4913 Color Points) for 8bit color depth is just about perfect.
If you go to a larger cube you will get 'noise' errors due to display/probe instability, which will greatly reduce the quality of the final result, unless you filter out the noise, which just gets you back to a smaller data set.
That's a very interesting comment, and I'll explain why:
Such a limitation is the characteristic of a profiling approach in which each measurement is taken literally, and the resulting device behavior model is a perfect fit to the measurement points. One sign that such an approach is likely being used is if it requires a regular grid of test points. A regular grid is very easy to create and turn into a profile, but easy is not necessarily optimal - a regular grid explores the RGB space inefficiently.
Using such a profiling scheme it's easy to understand why the use of highly repeatable and accurate instruments is of more importance that it otherwise would be (ie. Klein colorimeters ?)
A more mature profiling approach doesn't take each measurement literally, instead it makes an allowance for the inherent inaccuracy in each measurement, and uses the overall response of all the measurements to reduce noise and inaccuracy of the resulting device model. Although it may seem counter-intuitive, the resulting profile is more accurate than any single measurement. Another way of viewing this is that it is an approach that allows the averaging of different measurement points within the color space. This is not
equivalent to averaging multiple readings of the same test point, since it simultaneously explores the device behavior in more detail, and reduces measurement noise . One of the signs that such an approach may be being used is that it does not need a highly structured set of measurements, but can use a scattered data set.
The result is a profiling process that is more forgiving of inconsistency and noise in the instrument measurements, and allows the accuracy of more visually critical areas of the device response to be improved by structuring the measurement data set to sample these areas more densely.