Digital BW, The Print

--- In DigitalBlackandWhiteThePrint@yahoogroups.com, Steve Kale <stevekale@b...> 
wrote:
> I did a little more playing around with this and the following might provide
> some food for thought for RIP designers as they think about their
> "linearize" functions.
> 
> I was interested in what value for gamma would produce the least total error
> in print reflectance (XYZ_Y) vs that proposed by the CIELab model of
> vision's L*.  Someone with better stats skills could find a better measure
> of "fit" I am sure but this is a useful first observation.  I simply
> calculated the XYZ_Y value for each 8 bit value (and measured the difference
> between that and the XYZ_Y value that Lab would generate).
> 
> I assumed an average matte paper had a dynamic range of L*=16 to L*=96 and
> an average photo paper had a dynamic range of L*=3.3 to L*=96.  (The
> reference, Lab, of course goes from 0 to 100.)
> 
> For matte paper, if the RIP linearizes printed L* values from ink black to
> paper white (as they do today) the shape of the print reflection function
> (as measured by XYZ_Y) is best approximated (by this measure of fit) by a
> gamma of 2.05.  For photo paper this number is 2.38.  So these are the sorts
> of gamma implicit in the way we do things today.
> 
> I then asked what gamma would be the BEST fit to the benchmark, Lab, given
> the particular black and white points of the two papers.  This produced
> gammas of 2.38 for matte paper and 2.27 for photo paper, ie quite different
> from those produced by linearising L*.
> 
> All this of course assumes that getting as close to CIELab is the goal, ie
> that that model best represents the way we see.
> 
> Food for thought....
> 
> Steve

Hi Steve,

I think "what's optimal"   "what gamma?" or even "why a gamma curve at all"
ends up being somewhat of a circular argument.   I don't think anybody can
really argue that any one way is absolutely the best.   But most of the reason
for all this, is because for B&W the approach has been mostly a trial and
error to match the screen and print.   So if you happened to be using gamma 1.8
for your file you have one notion of what "middle gray" is but if you use
gamma 2.2 you have a different notion (middle gray is darker).  You then
tweaked the output method to match your notion.  However the ideal
thing is to allow anyone to use whatever they like in the file, profile what
your output device does and use color management to handle the conversion.

The only necessity is to make your output device reasonably well-behaved
so that the output profile was also reasonable.  Most OS print systems do
truncate to 8-bit somewhere in the flow so you don't want to loose too much.
Linearizing to L* is one way but linearizing to gamma 1.8 would also work.

Roy