Re: table resolution vs smoothness
Re: table resolution vs smoothness
- Subject: Re: table resolution vs smoothness
- From: Graeme Gill <email@hidden>
- Date: Tue, 26 Nov 2002 13:41:21 +1100
Roberto Michelena wrote:
>
Many profiling software developers say or imply in their documentation, that
>
there's a tradeoff between "color accuracy" and "smoothness (banding)"; if
>
you use a lot of patches and build a large number of gridpoints in the
>
profile, you get better numerical color accuracy but more noticeable banding
>
in smooth gradients. While if you build with few patches, you get smoother
>
images but less accurate.
These are often characteristics of the profiling algorithm. There
is also the issue of "noise" (measurement errors) in the patch readings,
causing apparent discontinuities in the devices behaviour.
Many interpolation algorithms allow a tradeoff to be made between
the profile exactly hitting the sample points and reproducing any
errors they contain, and reducing noise by smoothing transitions
in the profile, simultaneously reducing accuracy as defined
by the sample point values.
Often the limitations of 8 bit colorspaces are an issue in
trying to compensate for a devices character (particularly
with printers that have a lot of dot gain.)
>
That situation is terribly broken; if your original image (to be proofed)
>
has no banding in its final form (ie offset print), a "color accurate" proof
>
should't have any banding either. Unless "color accurate" from a numerical
>
model point of view, is not meaning "color accurate" from a human perception
>
point of view.
>
And unfortunately, the abovementioned warnings from manufacturers test true
>
in reality, which to my understanding can only mean our color appearance
>
model is terribly broken. Lab and XYZ have ripples, so to say.
Lab and XYZ don't have ripples. They are to any practical degree
continuous and smooth. Real device spaces on the other hand,
aren't always smooth, especially when viewed through discrete
samples that include measurement error. Color conversion mechanisms
(such as interpolation curves and table) are also not smooth. Most
are based on linear interpolation segments, and where the segments
meet is not smooth. Something like a matrix based Monitor or Input
profile can be smooth (because it defined by a set of simple
equations), but can't be super accurate, since it is simple,
and can't represent the detail of a real devices responses.
The exact tradeoff depends a lot on the profiling package. It is
possible to have very good accuracy with very good smoothness.
>
So when are we getting rid of those flawed spaces and replace them by
>
something that really represents human vision?
No time soon, I can assure you, since there's nothing much wrong with
them. Attempts to better represent human vision are being built
on top of such spaces - they are a foundation stone.
Graeme Gill.
_______________________________________________
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.