Alastair M. Robinson wrote:
Is there any special reason for the apparently random placement of the colour patches on the page? (It makes scanning the patches with the colormouse an "interesting" task, since with no software support, the XYZ values have to be added to the .ti3 file by hand...)
The randomization is there for a number of reasons, including averaging out device non-uniformity, recognizing strips, and making sure there is enough contrast between patches in a strip for the instrument to recognize the patch boundaries. It can be turned of by using the -r flag to printtarg.
Does Argyll have to generate the target itself, or can the patches be "hand-picked" or augmented? I ask because in my experiments last night I got a remarkably good result with a mere 60 patches - much smoother than either Prism or Print Profiler.
You can use any target you like really, you just have to get the final device and instrument data into a .ti3 file.
What I'd like to do is use a sparse scattering of patches, and then augment this in the "difficult" areas, adding some extra patches down the grey axis and in typical skin tones.
It's not really set up to do that kind of thing easily, since I was aiming at strip or XY instruments rather than one patch at a time type instruments. Don Bone had a interesting approach to that kind of thing, generating new test patches based on previously read patches, but it's not something I've pursued.
Finally, with both Argyll and Print Profiler (and to a much lesser extent, Prism as well), images printed using the generated profile generally have very good hue and saturation, but need further gamma correction to get the luminance looking "right" - in terms of Gutenprint controls, Composite Gamma needs to be increased to 1.1 - 1.15 or so. Any idea what's causing this?
It could be anything really. Check your source profile (is it really accurate for the material you are printing ?), intent (are you creating a gamut mapped perceptual intent table and using it ?) and viewing conditions (are they appropriate ?).
It depends how general Argyll's model is. I'm assuming the fact that it can adjust the patches based on the non-linearity measured in an existing profile means that the sampled data points don't have to be equally spread. The questions is, do they have to be some kind of n-dimensional grid, or can they truly be scattered data points?
Argyll's profiling copes with scattered sample data. There are some minimum requirements, but it should produce something with any sort of reasonable sample set.
For example, to create the contents for the table in the .ti1 file to augment patches along the gray axis for an RGB device you could do something like
$ perl -e 'for $i (20..99) { print "$i 0 0\n" }' | xicclu -pl -ir -fif profile.icc | awk '{ print $8, $9, $10 }' | icclu -px -ir -ff profile.icc | awk '{ print ++i, 100*$1, 100*$2, 100*$3, 100*$8,100*$9,100*$10 }'
Excel is really the tool alternative for someone not familiar with perl, awk etc.
Generally, one should check the minimal patch size specified by the instrument manufacturer. Accuracy may suffer if the patch is too small, even if the instrument still seems to just fit on the patch.
Right, backscatter through the media can cause color outside the aperture to influence the measured color.
Yep, those colours aren't easy to get right. I've done some work on the curve-based colour correction used in Gutenprint - and the real complication is that the printer's Cyan ink, and more specifically, the dark cyans that a naive RGB->CMYK conversion produces are *way* outside the monitor's gamut. It's very hard to create a simple model to compensate for that in a way that doesn't also desaturate the blues. Sounds like the driver you tried had similar issues...
If you're in a position to implement your own RGB/CMY to CMYK separation, the argyll can generate a pretty good one using the fakeCMY utility to create a cmy .ti3 from a cmyk profile, and then linking them to form a CMY to CMYK device link separation.
My goal is only to have to do one pass though, because I want to be able to profile without having physical access to the printers. My goal, BTW, is "looks good", not "colorimetrically perfect" - the ad-hoc curve-based correction currently used in Gutenprint is very difficult to get right - especially for skin tones, hence my special interest in those colours...
The default "full spread" algorithm is ideal for this sort of purpose.
Graeme Gill.
Graeme Gill.
