> --- Ursprüngliche Nachricht --- > Von: Graeme Gill <graeme@xxxxxxxxxxxxx> > > I'm guessing that instrument errors will be most significant at > the dark end, CCD sensors tend to have (in linear space) two noise components, an approx. constant amount of dark noise, and a 2nd noise component which grows approximately linear with signal level. At the dark end, the noise floor dominates the error, but at the light end, the 2nd noise component is dominant (btw, I mean absolute amount of noise, not the S/N ratio). But I don't know whether the same applies to photo diode sensors, maybe they avoid (or reduce) the 2nd component? > Given I was struggling to tame the complexities of determining > good smoothing factors for combinations of different dimensions, > number of sample points, and uncertainty level, That's indeed a complex and not easy task. Eventuelly, I guess it will always necessary to fine-tune the smoothness for each particular data set, but having a good, reasonable default starting point is IMO indeed very valuable. > I decided not to further complicate things at that time. Given the large uncertainty of the modeled noise, it likely does not matter, if only a single weight is used per data point. Btw, I could also imagine to scale the data point weights derived from the noise model additionally by a weighting factor derived from CIE94 dE weights, computed from the data point's Lab value, resulting in lower weights for high-chroma colors. From the pure mathematical fitting point of view, this likely does not make sense in order to predict the mean device behaviour with respect to Lab space as close as possible, but from a perceptual perspective, it probabaly does, and might increase the resulting smoothness of the splines further, without reducing the perceived accuracy significantly. Regards, Gerhard -- Gerhard Fuernkranz nospam456@xxxxxx Lust, ein paar Euro nebenbei zu verdienen? Ohne Kosten, ohne Risiko! Satte Provisionen für GMX Partner: http://www.gmx.net/de/go/partner