Graeme Gill wrote:
[...] One of the issues is that the color as seen by an RGB sensor in a camera is the product of the spectral characteristics of the illuminant, the spectral characteristics of the subjects, and the spectral sensitivities of the camera sensor. So a test chart may not be a good substitute for the subjects you want to photograph if it has substantially different spectral reflectance characteristics
In my experience, the metamerism failures Graeme mentioned are the key problem. The more the spectral characteristics of the camera/scanner sensor differs from the Standard Observer and the more the spectral characteristics of the test chart differs from the subject, the less you can expect from a profile.
In practice, you have to face with a multiple of the errors than those for the chart itself:
e.g. for a profile made from a photographic chart (dcam) "profile" reports: peak err = 5.445740, avg err = 1.118582 verification with a different chart (laser print): peak err = 18.542958, avg err = 7.665851profile made from several charts (inkjet- and laser prints, Colorchecker DC, photographic chart, ...):
peak err = 13.438882, avg err = 1.892351 verification with laser print as above: peak err = 11.950804, avg err = 3.576600=> you can improve the "overall" result slightly by using several (different) charts, but this will not solve the key problem (metamerism failures)
=> causal solutions:1) better sensor (spectral sensitivities must satisfy the Luther condition) => practical limitations
2) chart "tailored" to the subject (same pigments / spectral characteristics)
3) Multispectral approach (more than 3 sensors) Klaus