<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=ISO-8859-1" http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
Brian and Don and anyone else interested,<br>
I am not sure whether this has been published in this form, although
the idea has been bruited about in other contexts. It is certainly
something that I have been working on in my sputtering and
lackadaisical fashion.<br>
The gist is that the placement of glyphs and indeed any other visual
object should be dictated not by a computational geometry but by some
measure of importance and value. In other words, equal spaced glyphs
and contours are not only lazy but also misleading.<br>
What I would recommend is that glyphs be placed such that they lie in
the center of a region that contains a user-specified amount of
"stuff". For the vis scientist, this means doing a voronoi tesselation
based on the "stuff" field. For the flow buff, the obvious choice of
"stuff" is variance, with a close second to flux. <br>
Procedure (I would say flowchart, but the pun police are everywhere
and I have a reputation to project)<br>
1) select regions devoid of discontinuities<br>
2) within that region <br>
for every original sample point, compute the distance
weighted variance to nearby points<br>
(clever people would use multi pole methods here) computing
Sij for i>j only. The fun of <br>
course is to find a compelling version of the distance
between 2 vectors as a single scalar.<br>
For incompressible flow, the dot product works. Otherwise,
you have more options in<br>
blending magnitude changes with changes in direction. <br>
3) do voronoi tesselation on S, or smoothed version thereof<br>
4) place glyph at center of polygon <br>
I can root about in my files and look for slides from last year's
DoECGF forum if there is call for pictures. <br>
<br>
Alternatively you can compute a LIC texture and then decimate at a
user specified interval. This is easy but morally unsatisfying. And,
unless you have taken the trouble to build in Jean Favre's
implementation of LIC (PVLEA), <<i></i><a
href="http://www.cscs.ch/%7Ejfavre/Projects/vtkLEA/vtklea.htm">http://www.cscs.ch/~jfavre/Projects/vtkLEA/vtklea.htm</a><i></i>><i>
, </i>this is more work for you and the computer than can possibly be
justified. <br>
<br>
Richard Strelitz, Visualization Team<br>
Advanced Computing Lab (CCS-1), LANL<br>
<br>
<br>
</body>
</html>