[argyllcms] Re: Determining proper error value for -r

Roberto Michelena wrote: 
that's what I was thinking just now; without all the math, of course
:) but the basic idea:
- you either repeat patches in the target (a new targen option),
making it 8,000 instead of 1,000, or just measure 8 prints of the same
1,000 patch target.


That's not what I would recommend. If you're going to print 8000
patches, you're better off using them to explore the color space
in more detail, not just repeating the same test patches.

- a new column is added to the ti3, "patch reliability" or something like that 

It's something that Gerhard and I talked about a little while ago,
and technically it's not too hard to assign a "reliability" column
and to use it appropriately, since the mechanism is in place to
do this, but I haven't pursued this particular line, simply
because I'm not yet convinced that it will be worthwhile
improvement (and there are lots of other things to work on!).

To be practical, you need a way of estimating the variance of
each patch, without printing a patch value more than once.
There are ways to do this, though they may be impractically slow,
and may not be accurate enough:

For each test point {
        Remove the point from the set
        Generate the (lightly smoothed) interpolation
        calculate the difference between the point and
        the interpolation
        replace the point in the set
}

A variation on this idea would be to remove half the points,
and calculate the error for each of those to the interpolation
from the other half, then swap the halves and repeat.
You could do this several times with different random halves.

Basically, I expect this is all a bit too much for most
people. I expect most uses to simply use the default.
The profile -r option was for those who want to experiment or
have an unusual situation they are dealing with. Having an
extra, more complicated, slower option seemed to me to
be something that would rarely get used, and confuse
everyone else :-)

But if it looked like it would make a noticeable difference,
I'd be happy to add something like this.

Graeme Gill.


Other related posts: