Re: linearization - luminance, chroma or density?
Re: linearization - luminance, chroma or density?
- Subject: Re: linearization - luminance, chroma or density?
- From: "Roberto Michelena" <email@hidden>
- Date: Tue, 23 Jan 2007 17:08:55 -0500
I guess it's important to clarify what are the goals of
'linearization' and 'ink limiting' (usually both things are merged
into one) when it comes to inkjet.
1) to optimize the ink/media interaction (for a certain set of driver
parameters): find appropriate ink limits, both per channel and total,
so that ink does not smear of form pools or fills-in hairlines.
2) to optimize the use of the available gamut: if there's no
difference from 80 to 100%, you're throwing away 20% of your available
encoding values. So it's a matter of having the encoding values have a
better correspondence with the printed result (response values).
3) to facilitate the job of the profiling software: basically same as
above, but here the consideration of limiting the 'hooking' whenever
possible also comes into play. Also if you have much more gamut than
you need (for example for simulating a press), limiting it further
(per channel limits) optimizes even more the usage of the profile's
gridpoints.
4) to establish a known state to which we could revert (after drifting
happens from ink, media or environment variations) by the process of
"relinearization"
5) as a variation of the above, to establish a known state to which we
could bring other individual printers of the same make/family using
the same ink and media.
Now my take on these points:
1) I've seen no automated system do this reliably. EFI attempts to do
this together with (2) and (3), and ends up with a severely limited
gamut. The only reliable way seems to be by eye.
2 & 3) No way to separate these goals: there are several values on
which you could base your linearization, be it density, XYZ values,
CIE-based density (XYZ density), L*, Chroma, etc.; I personally have
toyed with the 'segment the length into equal dE steps" concept such
as Oskar proposes; Graeme also proposed at some point something like
"dE from paper white" which also sounds interesting, specially for
limiting.
Some linearization approaches are obviously flawed. Chroma, for
example, can give you trouble when it crosses over zero (yellow ink on
bluish media, cyan ink on yellowish media); L* seems nonsense for
yellow. Traditional density (status E or T) is somewhat noisy compared
to CIE-based density, and also has the problem of being targeted to
specific hues (offset inks) than might not be those of your device.
Then as it's narrow-band, when the hue shifts (hooks) in and out of
the sensitivity band, the density jumps up or down abruptly. So for
density I'd rather use CIE-based density (XYZ).
The only reliable way I've found to determine the ink limits to be use
(after the 'actual printability' limiting if needed) has been to print
a linearization target (primaries ramps) and graph it in ColorThink
along with the desired simulation profile (offset for example). Then I
can see more or less clearly what is the minimum ink I need on each
channel for achieving the desired profile.
But it's tricky because of the secondaries. It also becomes clear that
"max chroma" doesn't cut it in many cases, you need the extra darkness
provided by the hook after that.
So I've thought of making a profiling chart with few values under 70%,
(maybe only 10% and 50% for example) and then a lot over 70% (each 5%
step if possible); so as to map the gamut boundaries clearly. But it's
overkill right now, the ramp graphs serve me well.
One disadvantage I can see with Oskar's proposal of 'equal dE segments
along the colorant curve' (which is also my favorite), is that by
doing that, the hook (which is curved) is sparsely sampled, which
leads to poor modeling of it's geometry because of the spacing of the
profile's gridpoints. I think maybe some adjustment to that 'segment
size' could be made based on curvature radius.
4) oh, relinearization. Why does it suck so much in so many cases?
really. Many times it can't get you back to less than 3dE. So many
people just revert to making a new profile.
The main culprit, on my opinion, is that most RIP manufacturers use a
fixed blending between light and dark inks. Some of them because they
choose to use Epson's own driver API to inject CMYK contone, and have
Epson's code do the dithering and the light ink integration (in a
fixed way).
But the fact is, when you change your light cyan cartridge or your
dark cyan cartridge, the composite 'cyan' color changes not only the
reach of the curve (length, max chroma, whatever) but also the shape
(hook shape or start point) and then there's no algorithm that would
fit the new response over the old curve. The only way is to recompute
the blend, in other words, to linearize the light and dark inks
separately. ProofMaster does this, and I don't know who else might.
5) of course, same as the above. Also to note in both (4) and (5) is
that overprint behavior of the inks might also need to be corrected
(because of who knows what, maybe printhead alignment drifts); GMG and
ORIS do this by way of their "calibration to a reference printer", but
after failing miserably in the 'relinearization' part because of the
dark/light issue, it's a longer procedure than what it should be.
Maybe some 'secondary color linearization' procedure could be
implemented to produce an abstract profile to be inserted in the
middle of the chain, who knows.
On all these IPA and Wuppertal shotouts, a very important statistic is
missing: after having achieved those impressive 0.002 dE, raise the
room temperature by some degrees, change a couple of ink cartridges
(preferably buy them six months earlier in a neighbor country and
have them fly by courier), put in a new box of the same media (bought
also three months earlier and stored in an office shelf).
Measure the new dE (huge, I bet!), and then apply the linearization
routines of the respective software, and measure again.
That's some data I'd like to see!
best regards,
-- Roberto Michelena
Infinitek
Lima, Peru
_______________________________________________
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