sierra@das.llnl.gov (Frankie Sierra) (03/06/90)
This message is to report an error-diffusion filter which is more
simple to compute than Floyd-Steinberg, but produce the same per-
formance as this standard filter.
There have been many people confused about a filter attributed to
Floyd and Steinberg, but whose origins are really unknown. To sim-
plify this discussion I will refer to this filter as the False-Floyd-
Steinberg (FalseFS).
The FalseFS filter is described as:
* 3
3 2 (over 8)
This filter tends to lock in several patterns and artifacts for a wide
range of luminance spectrum bands. The Floyd and Steinberg filter
is more smoother in its transitions.
The real Floyd and Steinberg filter (F&S) was introduced in a paper by
its authors (Floyd, R. W. and L. Steinberg, "An Adaptive Algorithm for
Spatial Gray Scale.", SID International Symposium Digest of Technical
Papers, vol 1975m, pp. 36-37).
The Floyd and Steinberg filter is known as:
* 7
3 5 1 (over 16)
I have been playing for more than a year on error diffusion filter
design. And had already synthesized a bunch of filters with per-
formance close to the Jarvis, Judice and Ninke (1976) filter but
with less computational effort (If there is interest I could post
a summary of these filters). Just recently, in my quest for re-
ducing the filter to a minimum, I found out a filter with per-
formance similar to the Floyd and Steinberg filter, but less
computationally complex.
This filter, which I enjoy calling the Sierra-Lite, is:
* 2
1 1 (over 4)
This is the one I currently use when I want speed. However, I rather
use my other filters, when quality is of greater importance (the
difference being 25% to 40% more time for a 640 X 480 X 8-bits images).
Frankie Sierra
sierra@das.llnl.gov
76356.2254@compuserve.com