On Nov 26, 2012, at 7:33 PM, Graeme Gill wrote: > Iliah Borg wrote: >> Profiles are best applied to demosaicked data, and demosaicking needs to be >> run after >> white balance - given we want quality of course, not just speed. > > I guess I don't understand that conclusion. Is this due to noise > considerations ? Noise is sure one of the reasons. Demosaicking tends to skew colours as well. > Or is this to allow the demosaicking to better > synthesise detail from one color channel to the other ? To avoid demosaicking artifacts like false vector directions and labyrinths data is better be normalized in all channels before demosicking. We do white balance and demosaicking in full floating, while profile application usually results in fixed. Only matrix profiles can be easily computed and applied in floating point without re-writing CMMs. Very small difference in raw data in shadows can constitute a substantial difference in exposure; and in a lot of cases shots are underexposed 2 or 3 stops to hold highlights and even more if ISO is high. In processing those shadows are pushed and we need to be very careful to provide reasonable image data in shadows. We tried different ways, spent 2 years on that, and decided on the workflow we like the best quality-wize. -- Iliah Borg ib@xxxxxxxxxxx