[argyllcms] Re: Culling outliers from .ti3 input data before camera profiling
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Tue, 10 Jul 2012 22:17:27 +1000
Stephen T wrote:
Hi,
> ColorChecker Classic patches). All errors are dE2000 reported by colprof.
> Note the large blue
> errors:
>
> 140 patches: R = 4.1, G = 3.9, B = 8.7, av24 = 2.8, pk24 = 8.7 (blue) 139
> patches: R = 3.8, G =
> 3.1, B = 8.1, av24 = 2.7, pk24 = 8.1 (blue) 24 patches: R = 3.5, G = 2.9, B =
> 4.1, av24 = 2.7,
> pk24 = 6.8 (orange-yellow)
> Secondly, patches that fit rather poorly (outliers) may have undue leverage
> on the results.
The question to ask, is why there are outliers. Input devices are generally
quite additive since there are few mechanisms for colors to mix non-linearly,
so in principle a matrix model should work well.
It's been suggested by one of the people on this list that the main source
of such errors is flare. A first order correction for flare is to have
a light trap in the image, and subtract that black from all the
other values. I'm not sure how far this goes in explaining such errors,
but it might be a direction to explore.
Graeme Gill.
Other related posts:
