On 11/27/04 12:55 PM, "Graeme Gill" <email@hidden> wrote:
> If you're trying to improve the accuracy of the B2A table, then sure,
> you can fiddle with the grid values to improve the result for a certain
> set of test values, but since you aren't examining the changes for all
> the other in-gamut values you're not testing, the overall B2A table
> accuracy is likely to get worse, even as the results for your test set
> improve !
The important thing in this scenario is that the test values are the
Of course you can't test all the gridpoints (17x17x17= 4,913 and 33x33x33 =
35,937) ; but if you subsample the grid (for example 9x9x9 = 729), for those
points you did not test you can interpolate the corrections ("moves") done
on the points you indeed did test. As long as you keep using the same
interpolation algorithm as you used to build the profile, it's consistency
(smoothness) will be preserved. Unless the errors were induced by
measurement noise (which is easily eliminated by averaging in the first
place), there's no reason to assume the accuracy will suffer.
And to make it better, the second iteration you can make in the points that
were untested in the first one. You could asign an "inertia" to each
gridpoint, beginning at zero, and increasing everytime it has been tested
(and thus moved) directly. That way each the second iteration has less
effect on the previously-corrected points. And so on.
Hell, I'm just throwing untested ideas. But it would seem it's doable and
beneficial, it's just a matter of defining the right algorithm and tuning
it. It's a fact that in all fields of science, measure-and-adjust iteration
helps tune the result of a not-perfectly-fitting model.
-- Roberto Michelena
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