For anyone who considers this, don't worry that the camera throws an error message regarding the image, tell it to use the image anyway. When I first read about this method the suggestion was to use a completely overexposed image, and that is the way I have done it. The reasoning was that the dark frame may well have some noise, and if that noise was bigger in one channel over another it could skew the result. With a modern sensor at ISO100 and using the fastest shutter speed you have, any possible noise is going to be very low indeed. So I don't think it makes any practical difference for most cameras.
On a secondary note for Canon users if you have a supported camera, and are willing to install the software, Magic Lantern dose allow the use of a RAW histogram. I have used it on my 50D, but it wasn't really very useful as the displayed histogram was very small. It may be better on one of the supported newer bodies.
The problem with UniWB (I have been a big promoter on the net: http://www.guillermoluijk.com/tutorial/uniwb/index_en.htm ), is that even if WB is cancelled, the output colour profile conversion will take place. This conversion from RAW values to RGB values in sRGB/Adobe RGB spaces is performed through a matrix linear combination meaning that what you have in the end in the three channels is a linear combination of them all:
R_srgb = R_raw * k11 + G_raw * k12 + B_raw * k13
G_srgb = R_raw * k21 + G_raw * k22 + B_raw * k23
B_srgb = R_raw * k31 + G_raw * k32 + B_raw * k33
So saturation false positives and false negatives can occur (using a 0..1 scale where 1=clipped):
- Saturation false positive: e.g. G_raw<1 could be unclipped and G_srgb = R_raw * k21 + G_raw * k22 + B_raw * k23>=1 be clipped
- Saturation false negative: e.g. G_raw=1 could be clipped but G_srgb = R_raw * k21 + G_raw * k22 + B_raw * k23<1 be unclipped
Even if there is an obvious strong correlation between the clipped/unclipped condition of the three channels in the RAW vs sRGB/Adobe RGB domains (G_raw <-> G_srgb...), this means that UniWB cannot be perfect as a genuine RAW histogram would be.
Regards
