Gerhard wrote:
> And the partial derivatives dE/dRGB should not be too
> steep, since the RGB numbers are not floating point, but quantized 8-bit
> numbers (I think, it's not possible to feed 16-bit RGB into a Windows
> GDI driver, is it?).
Not sure about the Windows GDI. (nor do I know whether my printer
hardware and/or driver would be able to make use of the extra
precision anyway)
> Of course you also need enough patches to be able characterize a strong
> non-linearity accurately enough. A multi-pass approach, using "targen -c
> preliminary_profile.icc ..."
Yes, well, as I said, I was told that this functionaliy hasn't been proven
yet. :) Have you actually tried it, and if so, do you think it's working
well?
> for the 2nd and all subsequent passes is
> likely helpful for strongly non-linear devices as well, since with the
> -c option, targen will place more points in highly curved areas of the
> color space, and will distribute the points according to perceptuall
> criteria within the color space, instead of distributing them evenly in
> device space.
Yes, I understand. I have tried a very simplistic thing along these lines
with another profiler, and that was to create a preliminary profile,
determine the greyscale RGB curves required to linearise the greyscale
(by converting a greyscale wedge in a well behaved working space to
the printer profile), and then applying these linearisation curves to
the target data, and reprofiling. (Once the profile was created, I
put the linearisation curves into the profile by manually editing the
profile in a hex editor.) I think you've probably read all this
a while back in sci.engr.color - sorry for repeating myself. :)
Anyway, I felt that this did produce smoother
gradients in my very limited testing. Given that Argyll can use
so many patches in a target, the "-c" option seems less important
than it might otherwise me. However, if -c works well, then it would be
good to be able to use smaller targets, of course. :)
Greg.
Regards,
Gerhard
