We have added ICtCp as a dE measurement within LightSpace... but, our view is it is not at all good as a way to define dE.
Here's some background info on our thinking.
dE1976 and dE2000 are based on L*a*b* and are a measure of perceptual uniformity. This is not the same as JND (Just Noticeable Difference). From historical published work, it has been concluded that the average dE2000 that represented a JND is 2.3 (there is a ref on Wikipedia), and that is the value we use in LightSpace.
But, the real value of dE2000 for a JND is very dependent on the actual colour, as the L*a*b* is not completely perceptually uniform, just LOCALLY uniform.
However, that would be very difficult to show in a graph – hence our use of a 2.3 JND target line.
Dolby have concluded that a correctly calculated ICtCp value needs to be multiplied by 240 in order for it to have a similar magnitude to a comparable dE2000 value. (And we do tend to think this is about right as well, but have absolutely no idea why any absolute measured value should require a multiplier to make it 'viable'...).
But, Dolby also say that a dE2000 of 3 is about a JND! We say its 2.3, based on standard industry acceptance, but Dolby say 3? This may be experimentally true, but we are not confident that it is in any way correct. (But this fact has no bearing on the LS implementation – it is just a huge difference, and as we know most calibrations aim for about 1 or below dE.) So, to target a JND, the ICtCp value needs to be multiplied by 730 (240 x 3). As we totally disagree with a value of 3 for a JND we have not implemented this multiplier.
So, this is what we have done within LightSpace:
We have added RED lines to the dE and dE-Dist graphs. This is dE_ICtCp multiplied by 240, to scal to be compatible with dE2000. In this way the graphs are comparable as you look at them. All text values of dE_ICtCp are absolute, so look way lower than dE2000.
We are still not sure if this is exactly what real users will want, and we are not sure of ICtCp's viability as a dE measurement? We are most definitely not fans of dE_ICtCp... It is very poorly documented, still changing, and in no way a standard. It is way out of scale when compared to dE2000/dE1976. Worst of all it in no way takes account of the white reference, which is an integral part of L*a*b* and so dE2000/1976.
Basically, in our view, and until it can be proved otherwise, ICtCp is not a good metric to verify calibration accuracy against. (We are very open to input that explains we have misunderstood the technical data behind ICtCp, but nothing has been forthcoming as yet...)
Steve |