Path: utzoo!mnetor!uunet!tektronix!tekcae!kurtk
From: kurtk@tekcae.TEK.COM (Kurt Krueger)
Newsgroups: comp.graphics
Subject: Re: The 24-bit question.
Message-ID: <1528@tekcae.TEK.COM>
Date: 11 Mar 88 21:12:26 GMT
References: <927@xn.LL.MIT.EDU> <6818@iuvax.cs.indiana.edu>
Reply-To: kurtk@tekcae.UUCP (Kurt Krueger)
Organization: Tektronix, Inc., Beaverton, OR.
Lines: 20

It seems to me that the real question is how to represent a 24bit deep image
on a device that is only 8 bits deep without throwing away any more information
than necessary.  The solutions I've seen so far throw away more info than
needed, and indeed, can run into some severe problems if you run out of color
map before you run out of colors.

Some solutions that work very well are to divide up the available color map
as well as possible.  Something to remember is that the human eye has poor
spacial resolution with blue light, but very good color resolution (you can't
see much detail in blue areas but you can sure tell the difference between
shades of blue).  You then either dither to the available colors or use an
error difusion method.  With 256 shades of grey I've been able to error diffuse
images and not be able to see any color steps.  Having ~1300 pixels across, of
course, helps.

A warning on dither techniques:  If your output display is not "gamma corrected"
you will have to include this correction in your rendering code.  A device that
is not gamma corrected will not have equally perceived brightness changes
across the available range (i.e. you ask for 2x change and you actually get
3x or 1.5x).