Re: Media Testing for maclife.de
Re: Media Testing for maclife.de
- Subject: Re: Media Testing for maclife.de
- From: "edmund ronald" <email@hidden>
- Date: Thu, 11 Sep 2008 15:44:59 +0200
I have no issues with the reasoning here.
Confirming your reasoning by my own experience, I have found that
carefully made ICC profiles make users quite happy in many lighting
situations.
Hence, I would assume that interpolating between two profiles would
improve matters even more. This the assumption underlying the DNG
model, and I have no issues with the fact that it should be an
improvement over the single-measure matrix profile.
But now come the problems: There is no obvious reason to assume that
when the user hits mixed lights, daylight saving, neon tubes, mercury
vapor lighting, or other popular perversions, that profiles inferred
from smooth illuminants should still carry over here.
In other words, in ideal situations the old methods already work
pretty well, and in ugly situations the new ones are at loss too.
My own experience has been that a matrix profile computed on the fly
by a one line Matlab prorgram will supply better results in real-life
than any adjustments within a Raw converter. Luckily, Adobe now
supplies a user-level tool to compute these matrices.
> Well, it all depends. As I'm sure you realize, there are many, many (well, infinitely many) spectra that can map to the same correlated color temperature value. I have a standard household incandescent bulb here that measures 2850 K (pretty close to illuminant A) but also a Home Depot compact fluorescent "warm" bulb that also measures about 2850 K. They have completely different spectra and if you visually compare a ColorChecker (or any reasonable target) under the two lights, they will look very, very different. So, clearly a profile built based on one of those is going to have some difficulties with the other.
>
> There is the concept that if you shoot a target very, very carefully, with a single light, away from all other color influences, minimizing reflections, etc. then a profile built from that capture will work well under all lighting. That is largely true as long as the spectra of the test lighting is closely related to the spectra of the training lighting (i.e., the light used to shoot the target), and it holds true for many flavors of natural daylight, with its fairly smoothly-varying spectrum.
>
> If this was really true in general, however, we should expect nearly identical behavior when analyzing the white-balanced raw data (i.e., in camera-native linear coordinates). That is, shoot a raw file under one lighting condition, shoot the same target under another lighting condition, and the white-balanced raw values should be pretty close. Unfortunately, in general they are not. Some materials will be close, even nearly identical, but others will be very far off, with a very different balance in R/G and G/B. From this, it should be clear that a single profile (whose responsibility is to map the white-balanced camera-native RGB values to colorimetrically-meaningful values, such as XYZ or Lab D50, or RIMM) cannot work for both.
_______________________________________________
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