Re: Erroneous color transformations with Apple CMM
Re: Erroneous color transformations with Apple CMM
- Subject: Re: Erroneous color transformations with Apple CMM
- From: Klaus Karcher <email@hidden>
- Date: Tue, 26 Jan 2010 23:53:22 +0100
Marco Ugolini wrote:
Klaus Karcher wrote:
I know that the embedded profile in the source image is fairly
off-key and the image is over-exposed, but that's not the point.
Maybe it IS the point.
If you assign AdobeRGB to the image, for example, none of that
misbehavior occurs upon using the Apple CMM.
Maybe the "CMP Canon EOS 400D as" profile that is embedded in the
image file is non-ICC-compliant or somehow defective, and therefore
misbehaving, and for that reason it does become the issue here.
The only questionable point about the profile is that it contains
negative XYZ values. Apart from that it is fully ICC-compliant. Negative
values in matrix profiles are quite common (and useful) and normally
don't cause any problems.
The profiles I used in my numerical examples are fully ICC-compliant
(verified with Apples's Profile First Aid and SampleICC, a reference
framework released by the ICC).
By the way, the image looks badly overexposed only when the Canon
profile is assigned to it. When AdobeRGB, or sRGB, or ProPhoto RGB,
or one of the other standard working spaces is assigned to it, the
detail in the clouds becomes visible again (as it should be, since
RGB numbers in the low 200 range should not be represented as
featureless white), and no color drift occurs when switching to other
CMMs.
Of course you can get any result you want if you assing a different
profile, but as I already said: this is not the point.
... but OK -- if you demand for the whole storry: It's a small detail of
a photo with very high dynamic range. There are large areas in the
uncropped image that are rather under- than overexposed.
The image was processed in a linear workflow -- assigning anything else
but a profile with gamma 1 is simply inappropriate. The main part of the
image gets almost black if you assign a profile with gamma 2 or higher.
The profile was made from an underexposed shot while the photo is
overexposed.
I have chosen this image as it is a good visual demonstration for the
bugs I seemingly found.
If one concerns oneself with the numerical examples I posted to to the
developers list[1], one will find that there's not much wiggle room
left.
We are talking about matrix profiles and hence very simple and
unambiguous computable linear matrix transformations. Ironically the
largest errors occurred with floating point transformations where not
even rounding errors play a significant role.
In my numerical examples there is no need for fancy gamut mapping,
lookup table interpolation, not even for chromatic adaptation. Just
simple RGB <-> XYZ matrix transformations. I's all about pure,
unambiguous mathematics. The result can be only right or wrong.
Klaus
[1] sorry, one of the links was wrong in my last post:
<http://lists.apple.com/archives/Colorsync-dev/2008/Dec/msg00002.html>
<http://lists.apple.com/archives/Colorsync-dev/2008/Dec/msg00003.html>
_______________________________________________
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