> --- Ursprüngliche Nachricht --- > Von: Graeme Gill <graeme@xxxxxxxxxxxxx> > An: argyllcms@xxxxxxxxxxxxx > Betreff: [argyllcms] Re: FWA correction in XYZ space > Datum: Tue, 23 Aug 2005 09:48:16 +1000 > I guess it still leaves the issue of how to derive > the transform parameters from a minimal set of > measurements. Graeme, I've tested models with 3x3 matrix + offset, and 2nd order polynomials. The former model has only 4x3 and the latter only 10x3 model parameters, which could be dertermined from 4 or 10 trichromatic measurement pairs. However, since we're talking about data fitting, which eventually involves measurement errors and also a residual systematic model error, the "optimal" 4 or 10 patches should be picked to obtain best overall model accuracy, or - even better - a least squares regression with more than the minimum number of points needs to be done. As you see in my results, I have already tried to do the regression with only 100 full-spread points, and to check against 1700 points, and the error is nearly the same as when the regression is done with all 1700 points (IMO no statistically significant difference). Sure, 100 points are probably not yet a "minimal set of measurements" - I will also try the regression with even fewer points and check the results. Do you have any idea, how one could determine an "optimal" small subset of my measurements for the regression (without need to try e.g. 10^1700 combinations)? > It also tends to validate Gretag's approach, which (I have > been given the impression) is a XYZ_fwa_comp = f(XYZ) / > Lab_fwa_comp = f(Lab) type of approach. I don't know for sure, but I think Gretag's FWA compensation is also only available in conjunction with *spectral* measurements, is it? This would IMO suggest that they rather use an approach like yours (or even use your method?). 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