Hi,
Graeme Gill wrote:
I'm working on the project partly for my own satisfaction, and
Yep, I'm familiar with that motivation :)
partly on the basis it's easier to convince people with a working example, than by merely "telling them so". The latter approach understandably gets their backs up ...
It doesn't really change the problem, since the idea is that the source gamut is contracted and expanded to the PCS gamut, then the PCS gamut is contracted and possibly expanded to the destination gamut. If the PCS gamut was a very large one, and the source simply sat within it, then the destination mapping would contract the whole colorspace (on the basis of making the PCS gamut fit), reducing the colorfulness of the result dramatically. The suggested standard PCS gamut is a fairly large "printer like" one.
Yes, I do see the problem...
Yes it takes some time (about the same time as generating one of the B2A tables of a profile), but it's not so bad at medium resolution on a modern GHz processor. The end result is noticeably smoother and more accurate.
Very interesting. Time to book some time off work methinks :)
Very rare I would imagine.
All the best, -- Alastair M. Robinson