Re: Profiling Epson 5000
Re: Profiling Epson 5000
- Subject: Re: Profiling Epson 5000
- From: email@hidden (Bruce Fraser)
- Date: Fri, 3 Nov 2000 21:02:39 -0700
At 11:38 AM -0600 11/3/00, Bruce J. Lindbloom wrote:
>
Bruce Fraser wrote:
>
>in my experience, at least. once you get the RIP to produce
>
>reasonably linear output from the Epson (which is quite easy to do),
>
>it's quite profilable.
>
>
To which Glenn Kowalski replied:
>
> What needs to be done to have the 5000 produce linear output? And an
>
> even more basic question--what is linearization?
>
>
Glenn's question sounds obvious on the surface, but I think it is a really
>
good question. Linear means that if you plot the data on a piece of graph
>
paper, the points lie on a straight line. But the crux of Glenn's question
>
is "what does the data represent?".
>
>
If you linearize pixel value vs. reflectance you get one thing. If you
>
linearize pixel value vs. density, you get something completely different.
>
Technically, both situations are linear. So it is not enough to say that
>
something is linear. You must say that it is linear with respect to
>
_fill_in_the_blank_.
>
>
My opinion (and I'll probably get flamed for this) is that linearizing pixel
>
value vs. reflectance is a bad thing--it is the printing world equivalent of
>
using gamma = 1.0 in a monitor space or RGB reference space. You run the
>
risk of losing visually significant detail in the shadows to quantization.
>
For the past 10 or 12 years, I have been a proponent of using the CIE L*
>
function instead. Linearize everything to it.
>
>
Looking at it another way, it is fortunate that monitors have a "natural"
>
gamma of about 2.2. This behavior is close to the L* function. On the
>
printing side, most devices have "dot gain", which is actually helpful
>
because it gives the device a natural behavior which is closer to the L*
>
function than if it had no dot gain at all. FWIW.
I'm certainly not going to flame you for advocating linearizing to L*
-- it makes considerable sense. I would, however, point out that for
the application under discussion, it's probably a lot more important
that you linearize to something-dot area, density, reflectance, or
what have you-than that you linearize to any single one of them.
If you don't linearize before you profile, you generally wind up with
a lot of patches on the profiling target that are basically black.
The resulting profile usually displays some randomness in the shadows
as a result.
I've obtained good results linearizing dot area, and I've also
obtained good results linearizing to L*, but the latter gets
complicated when you have inks with more than 100% reflectance.
Profilers don't always handle the situation elegantly when a 10%
yellow dot has a higher luminance than paper white...
When I linearize to dot area, I usually use 2% increments in the 0-10
and 90-100 ranges. I haven't noticed any particular problems with
loss of shadow detail when I do that.
I've also experimented with linearizing CMY to dot area and K to L*.
So far the results are inconclusive but at least not noticeably worse
than using dot area. More work is needed...
I'm really, really sorry about BruceRGB. Had I known then what I know
now, I'd have called it something more impressive, like Fraser RGB
(1998). If you come up with a space that we both feel deserves the
BruceRGB monikker, I'll happily rename mine, sending press releases
to the NY Times, etc. Or you could always just go the Lindbloom RGB
(2000) route...
Bruce
--
email@hidden