Quote:>I am looking for references on multibit dithering (Ordered and/or Error

>Diffusion).

There is a very simple method for the special case where the

grey scale is to be quantized uniformly (equal steps):

Say you want to quantize to N levels, and that the input

is normalized to a min of 0 and a max of N-1. First quantize

the image to N-1 levels by rounding down, resulting in an

integer image ranging from 0 to N-2. (To insure that no

pixels get quantized to N-1 you have to tweak the normalization

so that the max is slightly less than N-1). Now subtract the

quantized image from the original, which results in a non-negative

error image ranging from 0 to 1. Apply your favorite binary

dithering algorithm to this error image. Now add the dithered

error image to the quantized (truncated) image, with the "on"

pixels in the dithered error image "promoting" selected pixels

in the quantized image to the next level.

I discussed this briefly in an SPIE paper from 1990 (of which I

have reprints). There are also other, more complex algorithms

which can deal with nonuniform quantization and arbitrary

color palettes, for which I cannot supply references off-hand.

For arbitrary color palettes, this usually boils down to

finding the "best" color; there are two halves to the problem:

the search (maybe it can be restricted), and deciding what

metric to use to decide which of two palette colors is "closer"

to the target color. General formulas for characterizing

perceptual color differences is an active area of psychophysical

research.

--

NASA/Ames Research Ctr., Mail Stop 262-2, Moffett Field CA, 94035

(415) 604-3745