Graeme Gill wrote:
Asman, Andrew J wrote:compared to the Gretag Spectrolino is consistently under 2. However, if I measure pages using a printer that has a larger gamut than the one used to create the profile, then the out-of-gamutcolors tend to produce extremely large max delta E value ranging anywhere up to 100 depending on the printers used.It's no surprise that extrapolation is much less accurate than interpolation, and that the further away from measured values youget, the worse the error.Is there any way to increase the accuracy of the profile when faced with out-of-gamut colors?I'm surprised to hear of delta E's as large as 100, unless you are significantly out of gamut.
I've also seen abruptly and dramatically increasing errors with Lab CLUT scanner profiles, most notably in the shadows.
I've written a tool to pre-calibrate and linearize scans to account for illumination and scanner gain/offset variations. It does a good job most of the time, but sometimes it sets the black point imprecisely for whatever reason, resulting in a slight clipping of the shadows.
When this is the case, it can happen that there are suddenly blazing shades of green, blue or red showing up in areas where one would expect nearly black. Most of the time, this phenomenon becomes effective not until the image has been transformed to the output space. Therefore I suspected gamut mapping issues in the output profile to be the culprit, but in the meantime I noticed that the problem vanishes when I assign a shaper+matrix profile as source profile.
You could try shaper/matrix, or gamma/matrix for instance, but you will probably be trading off a worse in gamut fit for a better out of gamut fit.
Yes, and this tradeoff is the reason why I prefer to use the Lab CLUT profile normally.
This is assuming that the scanner has an underlying behaviour that is close to a per channel compression curve...
I hoped to be able to conjoin the advantages of both types of profiles by using XYZ CLUT profiles (as the ICC spec permits to combine CLUTs with matrices in XYZ profiles), but even though my XYZ CLUT profile had the least self fit error, the results were disappointing: I got unacceptable banding and hue shifts in gradients :-(
I inspected the XYZ CLUT profile and noticed that the A2B tag contains the identity matrix.
Could XYZ CLUT profiles with "real" matrices be an option to improve scanner profiles?
Klaus
