[argyllcms] Re: Rendering intent?
- From: Graeme Gill <graeme@xxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Wed, 27 Nov 2019 15:35:28 +1100
Samuel Chia wrote:
Usually, they forgo way too much lightness to preserve some
saturation, in an attempt to make the shortest path into the output gamut.
The distance of the mapping depends on the metric space.
Typically colorimetric mappings weigh luminance distance more
heavily than chromatic distance, meaning that delta's in luminance
are less than deltas in chromanance.
For Perceptual, we have been misled to believe that the only possible
mapping is a non-linear compression type, and so the mapping is
"unpredictable" and thus completely different secret recipes from one
software to the next. In fact, there is nothing preventing the Perceptual
intent from being treated the same way as Colorimetric for in-gamut
colours, and when we open ourselves to this way of thinking, we can
consider something else that is much more important.
Any mapping that maps colors colorimetrically leaves no volume for mapping
out of gamut colors into gamut other than by clipping them. So by definition
a non-clipping (i.e. perceptual) gamut mapping must compromise in-gamut
absolute accuracy to improve the relative accuracy of colors between themselves.
I would like to correct the concept that the shortest movement in 3D space
is necessarily the best conversion when it comes to regular photographs. As
Florian has pointed out, lightness shifts are most noticeable while chroma
shifts are the least.
As mentioned above, this is typically weighed into the space metric.
When many people do gamut mapping comparisons, they like to use synthetic
gradients to evaluate the mapping of profiles for both Colorimeteric and
Perceptual tables, which gives a completely false impression of how well
they work for ordinary photographs, which normally vary significantly in
hue and lightness in a small region, not smoothly graduating from one tone
to another, one colour to another, and usually only have small regions of
out of gamut colours.
On the contrary, most gamut mapping development makes extensive use of
representative imagery. Synthetic images are typically only used to
evaluate technical aspects, such as smoothness or inconsistencies in the
mapping.
Thus, lightness-preserving gamut mapping is
significantly better at maintaining the appearance of the original image,
especially when out of gamut colours are concerned, to avoid loss of detail
due to excessive contrast reduction.
I suspect this depends a great deal on the aesthetic of the viewer. Someone
(say)
who deeply appreciates monochromatic photography may judge an image
much more by its luminance accuracy and gradation subtlety than someone
who normally views color images, and judges color impact as more significant.
Preserving relative contrast is always an aim in gamut mapping, but spatially
independent mappings such as those performed by ICC tools and similar,
are limited by this. Powerful gamut mappings aren't restricted to be
spatially independent, and can therefore use a much wider range of techniques
to maintain relative contrast.
Lightness-preserving gamut mapping is the key to getting successful optimal
results on relatively low contrast matte media as opposed to glossy media,
In your view. In my view, the rendering that you seem to prefer
have lost far too much vibrancy to be considered good renderings,
even though they may faithfully preserve luminance.
I any case, I suspect that really good gamut mappings cannot
be achieved by spatially independent analysis, which is all
that is available in the usual ICC workflow. You were
also resistant to using more advanced workflows, such as image
specific gamut mappings.
however, despite my many attempts at encouragement to Graeme to implement
this properly, including paying him an hourly fee and travelling to
Australia to work with him full time to get this correct and eliminate
communication misunderstandings over email, I have not been successful at
convincing him.
Dangling carrots only works for so long. Although I spent some
months attempting to understand what you were looking for,
either you failed to convey it, or I failed to grasp it.
In any case, there is a limit to how much time I can
spend on wild goose chases.
Beautiful, top-notch gamut mapping is the last great hurdle
of digital imaging yet to be breached for the reproduction of general
photographs of fine quality. Many still don't understand just how important
this is for the visual quality of their prints so it is rarely discussed,
understood, and thus there is little pressure on profiling software makers
to make radical changes to their mapping.
Yes, implementing powerful gamut mapping would be a fun project
(along with a long list of others), but unfortunately I'm certainly
not in a position to devote time to it.
You are mistaken in thinking this a color profiling issue though.
Gamut mapping at the profiling level is mainly about automatically
dealing with different output referred colorspaces.
Adapting input referred images to output referred spaces is
something quite distict, and a different toolkit and workflow is
desirable.
Most cameras and raw image processors incorporate
various automated rendering models to produce fast, output
ready images, but naturally such renderings won't
satisfy the demanding photographer. This is the point
where gamut mapping from input referred to output referred
imagery should be applied though.
Graeme Gill.
Other related posts:
