Re: Big problems using Bill Atkinssons new Eye-1 profiles
Re: Big problems using Bill Atkinssons new Eye-1 profiles
- Subject: Re: Big problems using Bill Atkinssons new Eye-1 profiles
- From: Graeme Gill <email@hidden>
- Date: Mon, 06 Feb 2006 13:41:38 +1100
Jim Rich wrote:
And if there are claims that a higher patch set does produce a better
result, is there hard scientific evidence to support that claim?
It doesn't take much thought to show that typical test charts are very
sparsely sampling the devices. A typical high quality RGB ICC A2B table
has 33^3 = 35937 grid locations that need populating. 918 patches amounts
to sampling 2.5% of the grid locations. Put another way, if the sample
patches were perfectly distributed in the color space (they are not),
each sample point represents approximately 780 cubic delta E of gamut
volume. (Assuming an RGB printer gamut volume of 719000 dE^3 in total
- the volume I measured on my Epson R1800 in RGB mode).
There's plenty of scope for devices behaviour to deviate from the
interpolated prediction due to such a sparse sampling. Now whether this
matters depends on the charts point distribution, the characteristics
of the device, and the end purpose. If all you're after is pretty pictures,
and the device isn't badly behaved anywhere in it's gamut, 918 might well
be more than enough.
I've done some rather extensive simulations of profiling accuracy vs.
number of samples and device & instrument uncertainty, and basically
the profiling accuracy keeps improving with more patches, no matter
how many patches you use (and I've simulated up to 1000000 for CMYK).
You do have to smooth the interpolation appropriately to match the
device & instrument uncertainty, and the latter causes there to
be a "knee" where you start to get diminishing returns.
If you don't smooth appropriately, then you can also reach a point
where more patches makes the accuracy worse too (the extra patches
are adding noise faster than they are adding extra device
behaviour information.) I would suspect that some profiling
systems fall into the latter sort of trap, and limit the return
on using more than a few thousand patches. [Of course, a few
thousand patches is often a practical limit in any case.]
In practical experience, using more than about 3000 patches
for profiling CMYK devices seemed (with the tools and devices
used at the time) to be at the point of diminishing returns.
I occasionally ran across issues with input profiles created
from around 1000 non-uniformly sampled points (ie. ECI chart)
not behaving at certain critical input colors, where augmenting
the point set with those critical colors, would pull the proof
into spec. (and the input devices were usually "well behaved"
printing presses or other screened devices).
Graeme Gill.
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden