Am 27.11.2012 17:08, schrieb Ben Goren:
On 2012-11-27, at 8:52 AM, Gerhard Fuernkranz wrote:I.e. the question is: Which (absolute) XYZ color would the patch need to have, in order that camera + raw converter would return non-clipped RGB of [1,1,1] ?Actually, I think the approach would be from the other end: what's the XYZ color that causes clipping at maximum RGB,
Interesing is the point, where clipping _starts_. And since there exist an infinite number of XYZ triples which would cause clipping to maximum RGB, it is better to approach the starting point from the other side, where the RGB to XYZ function is still bijective.
but just barely, and causes no clipping at all for RGB values just a smidge less? Again, empirically, the synthetic D50 white patch I manually added does exactly that: 1.000000 1.000000 1.000000 [RGB] -> Lut -> 100.000000 0.000090 -0.000032 [Lab] 0.999999 0.999999 0.999999 [RGB] -> Lut -> 99.999950 0.000090 -0.000047 [Lab] 0.999999 1.000000 1.000000 [RGB] -> Lut -> 99.999998 -0.000105 -0.000068 [Lab] 1.000000 0.999999 1.000000 [RGB] -> Lut -> 99.999939 0.000352 -0.000205 [Lab] 1.000000 1.000000 0.999999 [RGB] -> Lut -> 100.000000 0.000006 0.000164 [Lab] That seems to me to show the maximum possible preservation of highlights without compromising the white point, which I should think is the exact desired outcome.
Well this is eventually a tautology. Note, the brightest patch in the .ti3 file becomes the media white point, and when applying a profile with relative colorimetric intent, then the media white point will _always_ map to D50 in the PCS, regardless its absolute XYZ color. I.e. _any_ absolute XYZ color associated with the "WHT" patch will eventually map to white, if the profile is applied with relative colorimetric intent, and if the WHT patch is the britghtes one. But not any XYZ color of WHT is consistent with the other readings in the .ti3 file. Best Regards, Gerhard