Digital BW, The Print

--- In DigitalBlackandWhiteThePrint@yahoogroups.com, "Ernst 
Dinkla" <E.Dinkla@c...> wrote:
> To help me with the adaption of the Wasatch SoftRip for quad
> printing a friend wrote a small PC application to convert 8 bit
> and 16 bit greyscale files to plain 8 bit or 16 bit cmyk files.
> No separation is done just 4 copies of the greyscale gathered 
and
> given the C,M,Y,K, mark. Similar to what can be done with a
> PhotoShop action. He also added a choice of compression, 
LZW,
> ZIP, RLE or none. The ZIP compression is not accepted by the 
RIP
> but PhotoShop does, with that kind of CMYK files the size is 
then
> a bit smaller than the original greyscale file.
> I wonder whether it has any sense to make the application
> available on my site, with the LZW compression added it isn't
> wise to put in the file section of this list.
> 
> Following a small thread here (OPM in theory) on the 
conversion
> of 8 bit greyscale to 16 bit I asked him whether it could be
> added to the same application with the bit shifts as discussed
> here. He made some comments I like to add (my translation):
> 
> Quote:
> From the discussion I gather there are three variants:
> 
> i8  :   8 bits grey values (0-255)
> i16: 16 bits grey values (0-65535)
> 
> A) i16 = 256*i8
> B) i16 = 257*i8
> C) i16 = 256*i8 + random(0,256)
> 

Ernst,

I agree that B is the only correct one and btw, what PS does.

C is probably fairly harmless for most of the range,  but I think
at the end points you'll end up with some undesirable side
effects.   Consider white=(8bit)0, if that gets converted to
(16)bit0+random_number  its no longer white.  Its effect will
be to increase dMin by half a unit and decrease dMax by
half a unit.

Roy

> I consider B to be the only correct one. With that the complete 8
> bit intensity range is scaled to the complete 16 bits range.
> Software writers tend to do A that gives a small offset and a bit
> lower contrast. Some people in the discussion had high hopes 
for
> C. I don't get how you can get a better result by introducing a
> random falt of half the step value on purpose. Best guess is 
that
> it tries to generate a random dithering. I seems to me that the
> dithering process should be done on results obtained with 
exact
> computations.
> 
> If there's a desire to experiment I can add a 8-16 bit conversion
> option with a choice between B and C. Any thoughts on the 
random
> function in C are welcome. It can be a uniform distribution 
where
> all the numbers get an equal chance. A normal distribution 
may be
> better, the numbers further from the middle get a lower chance.
> 
> Gert-Jan
> 
> End of quote, translation by Ernst
> 
> 
> Ernst