[argyllcms] Re: Beta RGB As a Color Workspace

  • From: Chris Lilley <chris@xxxxxx>
  • To: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • Date: Thu, 17 Jul 2014 16:41:25 +0200

Hello Graeme,

Thursday, July 17, 2014, 1:38:13 PM, you wrote:

> Edwin Blenkinsopp wrote:
>> (Chris wrote)
>> 'Its possible to do a per-image colourspace analysis and to map the
>> convex hull of the image data to the destination colourspace; and to
>> have a smooth progression from "keep colours the same" near the middle
>> through "move them a little" near the edges and "clip less than 2% of
>> pixels" at the edges. but not in an ICC defined gamut mapping, I
>> think.'

>> Thats a great definition of whats needed for bulk conversion to web images
>> Any offers of any scripts thev have tried to achieve something like this?

> Well, the above is a pretty good description of what Argyll can do
> using a tiffgamut + collink + cctiff workflow, and making use of
> ICC based device profiles and device links. You do have to trade precision
> for speed a little - collink -g is going to faster than collink -G in creating
> the per image color transform, but -G is going to be more precise.

I wasn't aware that Argyll could do that, good to know.

I originally made the suggestion of that sort of algorithm in about
1992 when I was writing some teaching materials on colour for computer

It was implemented, as a demo, but a bit more crudely than I suggested
above. There were two variables, a percent of colours that were
allowed to be clipped and an ease-in factor (basically how close to
the destination gamut boundary you got before startin to scale colours
in towards the neutral axis.

The code no longer exists and was for old Unix workstations that
no-one uses anymore.

A useful addition would be to compute the aggregate hull for a set of
images, (for example, if they will be displayed together) so they al
end up using the same mapping and colours stay consistent. That could
fairly easily be done by just tiling them into one temporary mosaic

Best regards,
 Chris                            mailto:chris@xxxxxx

Other related posts: