re: Digital Camera: Raw Data + ICC profile = one very bad idea
re: Digital Camera: Raw Data + ICC profile = one very bad idea
- Subject: re: Digital Camera: Raw Data + ICC profile = one very bad idea
- From: email@hidden
- Date: Wed, 26 Feb 2003 14:18:50 EST
Tom,
As a retired color engineer and now a sometimes digital photographer helping
my pals (users and developers) implement color-managed workflows, I'd like to
respond to your post which, while generally true, looks primarily at only the
first step in concert with v2 ICC profiles.
<<Digital Camera: Raw Data + ICC profile = one very bad idea>>
That's only true if you stop right there, after creating a colorimetric
profile. Which, unfortunately, is where the third-party camera profiling
products do indeed appear to stop. I've profiled a lot of cameras with the
top packages and the results are barely colorimetric at best, as if the goal
of a camera profile was to act like a scanner. It most certainly is not. But
at least a relatively colorimetrically accurately reproduced image is a good
starting point for the professional who wants to work with such data. Most
amateurs, however, are not interested in this workflow for reasons you
described.
You nicely describe many of the reasons why "preferred" color reproduction
will not result from a scanner-like profile, but are generalizing a bit. I
think what you really mean is that a v2 colorimetric style ICC source
profile, is a very bad idea to reproduce the original scene. The camera
manufacturers seem to understand this to varying degrees (even if the
profiling package developers do not), that there is an important step after
the camera colorimetry and that is the perceptual rendering, preferences,
look of the image, or "bias" as you say. From my attendance at ICC and ISO
meetings, most color scientists don't at all "snicker" at the concept of
rendering, rather they take it very seriously indeed. In fact, that is a very
active area of both academic research and camera engineering.
So there are at least two parts to the equation. First getting a good
colorimetric characterization of the particular camera so that an accurate
scene-referred representation intermediary conceptually can be created, then
from this, render the image to a perceptual output-referred state for viewing
or printing which has the look or "bias" of a good preferred reproduction.
From my experience, the camera profiling packages do at best an OK job of the
camera colorimetric characterization and poorly address, if at all, the
perceptual rendering. So what we end up with at best is a decent
camera-specific profile, that has a colorimetric (dull, dead, flat, whatever)
style of rendering. On the other hand, the camera manufacturers might do an
OK generic perceptual rendering for their customer base but aren't accounting
for camera-specific colorimetry.
I think (hope?) that the situation will change now with the introduction of
the v4 ICC profile. Besides cleaning up things that are important to digital
photography (chromatic adaptation, better PCS definition, etc.), I think we
will see the use of multiple rendering intents in camera profiles. At the
very least, a camera profiling package could put a colorimetric style of
rendering in the colorimetric (duh) rendering intent and use the perceptual
rendering intent to do, uh, perceptual rendering. There are already v4
profiles floating around that do this, more or less, and the results are
quite remarkable.
That will go only so far with current "dumb" CMMs, though it should be better
than the current state of affairs. In the scenario just described, the same
basic perceptual rendering would be used with all images. In fact, the
optimum rendering is image-specific. And this is where smart CMMs will come
in. Perhaps using the colorimetric rendering intent to get back to scene
colorimetry the smart CMM can look at metadata and image statistics to do a
more optimal rendering. Some cameras already do something like this, to
varying degrees.
So I'd like to modify your proposition:
Raw Data + v2 colorimetric ICC profile = a mediocre idea but an OK starting
point
Raw Data + v4 ICC profile (with colorimetric and perceptual rendering
intents) = a very good starting point
Raw Data + v4 ICC profile + smart CMMs = where we wanna be next
Eric Walowit
Tahoe
_______________________________________________
colorsync-users mailing list | email@hidden
Help/Unsubscribe/Archives:
http://www.lists.apple.com/mailman/listinfo/colorsync-users
Do not post admin requests to the list. They will be ignored.