| Re: ISO DR definition - erratum [ QUOTE ]
Hi, John,
[ QUOTE ]
Most converters don't treat noise symmetrically. They just call the center of the bell-curve of noise (from the black pixels) black, or zero, so half of the noise gets clipped to 0 (all the noise that would be negative).
[/ QUOTE ]
Exactly.
[/ QUOTE ]
I have to wonder how many different approaches were played with. I see multiple possibilities, some of them are:
1) treat pixels below the blackpoint as non-existent, as if they were defective (interpolate them from the "good" ones)
2) demosaic first, and then clip the black - this would cause some averaging which would push more of the resultant values above 0 before clipping - this would brighten the shadows a bit, and may require a subsequent transfer curve to darken
3) put the whole bell-curve of noise in the image, but compress it without affecting other tonal ranges relative to black - this would need to know white-balance parameters before-hand to do correctly
4) a combination of 2&3
5) just clip the average black value - this is what I think most are currently doing - and for some cameras, I think it is done wrong (the 10D has a green cast in deep shadows - possibly from a poorly chosen blackpoint) |