Path: utzoo!yunexus!stpl!yunccn!nccnat!root
From: root@nccnat.UUCP (Paul Shields)
Newsgroups: comp.graphics
Subject: Re: FAST flood fill
Summary: far faster fills found -- References
Message-ID: <289@nccnat.UUCP>
Date: 5 May 88 01:40:47 GMT
Article-I.D.: nccnat.289
Posted: Wed May  4 21:40:47 1988
References: <1210@sbcs.sunysb.edu> <23807@ucbvax.BERKELEY.EDU> <288@nccnat.UUCP>
Organization: York University, Toronto Canada
Lines: 34

In article <288@nccnat.UUCP>, I write:
> There's a paper kicking around somewhere by someone at Berkeley, circa 1986, 
> analyzing all previously known fill algorithms and presenting a new one. 

Fishkin, Kenneth P., and Barsky, Brian A. 1985. "An Analysis and Algorithm 
	for Filling Progagation", Graphics Interface '85. 

The algorithm takes care of  U, S, and W turns. 

> A friend of mine at IBM has optimized this alogorithm, achieving high-
> performance fills for multiply-connected regions:
> 	- worst case 1.5 visits per pel;
> 	- best case 1.0 vists per pel;
> 	- avg = 1.05.  
> 
> For simply-connected regions, it's 1.0 visits per pel always.

He is, Ian Ameline, Development Analyst, IAD, IBM Canada Lab, 
TOROLAB2(AMELINE).  I have a drop-box here for him, ian@nccnat.UUCP. 

I believe his optimizations include improved handling of some types of
regions and use of memory management routines to reduce thrashing, since
Fishkin and Barsky also report avg 1.05 visits per pel, 1.5 worst case, 
but admit that for simply-connected regions it's not always optimimal.

Unfortunately, the (optimized) code is proprietary.  But I _may_ be able 
to convince him to hint at what hasn't been taken care of in the other 
algorithm.  In any case, you can see it at work in the IBM ImagEdit 
software if you're interested.

Disclaimer: I have no connection with IBM or any of its products. Please
do not take the above paragraph as an advertisment. 
-- 
Paul Shields, shields@yunccn.UUCP   If you think you have a subconscious,
or yunccn!nccnat!root               you have a software integration problem.