-------- Original-Nachricht -------- > Datum: Tue, 26 Aug 2008 13:18:31 +1000 > Von: Graeme Gill <graeme@xxxxxxxxxxxxx> > An: argyllcms@xxxxxxxxxxxxx > Betreff: [argyllcms] Re: Perceptual spread in targen for non-linearized > devices > The only explanation I can give is that the device characteristic is > a mapping between device and perceptual (PCS) space, so it's the > way these spaces are connected by the profile curves/surfaces that > is key, and that the spacing should not be too sparse in either > the domain or range. I't IMO not so surprising that a non-parametric regression with RSPL may result in overshoot (and thus reduced generalization accuracy for colors not in the training set) in regions where the independent variable (i.e. device space) is sampled too sparsely. And with perceptually evenly distributed points, a sparser sampling is to be expected in regions with flat perceptual/device gradients, where large changes of the device values result only in small perceptual changes. For well-linearized devices the risk is certainly lower and I'd not expect so much difference. But I'm wondering whether it may be beneficial to take the pre-linearization tables into account, so that the points get evenly distributed in the pre-linearized device space which serves as input to the RSPL, and not in the native device space? Regards, Gerhard -- Psssst! Schon das coole Video vom GMX MultiMessenger gesehen? Der Eine für Alle: http://www.gmx.net/de/go/messenger03