[argyllcms] Re: Dispcal - Inverting Jacobian failed?
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 22 May 2009 01:40:12 +1000
Gerhard Fuernkranz wrote:
just by the way, do you happen to know this one?
http://www.mat.univie.ac.at/~neum/ms/snobfit.pdf
http://solon.cma.univie.ac.at/software/snobfit/
Recently I stumbled over this paper, and what initially came into my
mind was that this algorithm seems to be taylored for problems like
dispcal, i.e. for minimizing a non-deterministic, expensive objective
function. It seems to be a bit complex and therefore expensive to
implement though [there is however a Matlab implementation available
freely for download, but I did not yet find some time to evaluate it].
Hi Gerhard,
no, I haven't come across this paper. I'll take a look at it,
but I imagine it might be complicated to implement outside of Matlab :-)
The current dispcal routine has some degree of robustness in the face
of bad behavior and noise in that it keeps a limited history
for each point, and reverts to a previous best point if things
go astray, but it probably isn't the last word in robustness or
performance. Keeping an updated Jacobian was an attempt to improve
performance.
in particular clipping at the high end seems to be very common among
different LCD display, when the contrast and/or RGB gain controls are
turned up too much. I think it may be helpful for the user, if the
dispcal functions
2) White point (Color temperature, R,G,B, Gain/Contrast)
3) White level (CRT: Gain/Contrast, LCD: Brightness/Backlight)
would give aid to detect a clipped, non-progressive output (e.g. if
dispcal would print a metric for the current amount of highlight
clipping in addition to the current deviation from the target).
The thought that crossed my mind was that it might be useful
to have an option to measure a number of points and plot the
resulting curve. Perhaps there could be a couple of selections
(ie. 8, 24, 64 measurements) so that the user can choose how
long they want to wait.
Thanks for pointing out the paper, and your other thoughts.
cheers,
Graeme Gill.
Other related posts: