edwin@cwi.nl (Edwin Blake) (03/06/91)
Soft sciences argue endlessly about terminology to the disgust of (us) physicists. To continue the tradition, this note is about a viewer- centered (centred?) approach to computer graphics -- psycho-graphics being too scientific/Hitchcock like -- one wants room for both art and science in graphics. The discussion was, as you recall, about physics not providing the sole underlying basis for graphics. A dismissive definition of realism in graphics is "like a photograph". In an overblown form this even becomes `the Turing test for computer graphics'. Shall we agree that perspective projection lies at the heart of such realism? This `realism' has grave problems as many others have noted, at least since Leonardo da Vinci, if not Plato. Perspective as you or I normally see it is very seldom correct since one inevitably has to unscramble an off-centre view (head-mounted displays or peepholes of anamorphic art are exceptional). Yet the realism of photographs is taken as unproblematic enough to become the definitive example of realism. Please note: this not an attack on physics, photographs or perspective or anything. I am trying to provide a lever for breaking loose fixed ideas. The fact is that that off-centre perspective view IS realistic, i.e., convinces us, the viewers, that it is a window on a scene in some environment. But it also enables us to adopt other encodings of perspective. Even with photography, alternative perspectives are available. David Hockney's joiners (composite photographs) provide a very good example. I reproduce an example of his work in my paper called "The natural flow of perspective: reformulating perspective projection for computer animation" Leonardo (1990), Vol 23, no 4, pp. 401-409 & Color plate A. I work these ideas out further in that. Seeing (perception) is more than just the retinal interface. A picture can quite legitimately directly address higher level cortical representations -- and in that sense one can get modern art which is more convincing (realistic) than `window'-art. Consider two planes flying overhead in parallel straight lines from one horizon to another. How do you `picture' their paths in your mind? Straight lines? Converging at one end only as in perspective?! Converging at both ends in some curvilinear fashion? (A related topic: what sort of metric does Cyberspace have?) Edwin Blake, edwin@cwi.nl Phone: +31 20 5924009 Centre for Mathematics and Computer Science (CWI) Department of Interactive Systems, Kruislaan 413, 1098 SJ Amsterdam, The Netherlands
eugene@nas.nasa.gov (Eugene N. Miya) (03/07/91)
On a long drive back from the Sierra, I pondered this topic more.
Perspective does have uses. This is how we make maps, topographic maps,
and other maps. Is there any image more pervasive and full of more
useful information than a map? I don't think so.
The perspective of two images creates a radial displacement
(photogrammetry term) which our brains fuse together to comprehend
depth. We pick up other cues (e.g., over lap, time and age).
People spend hours pouring over images like these to create topo maps.
And look what we get from them: a little model of the world in orthogonal,
synopic view (these big words from my remote sensing classes). You can
adding the distances, figure out how long the drive tapes, the symbols
in the legend can convey non-geometric information. Most photographs or
synthetic images can't approach the information content of a good map.
With a computer and a good map, one could determine the volume of material
ejected by Mt. St. Helens. It's the quality of the model which is important.
"Welcome Mr. Beatty to the Parallax Corporation."
-- The Parallax View
--e. nobuo utsunomiya, NASA Ames Research Center, eugene@orville.nas.nasa.gov
{uunet,mailrus,other gateways}!ames!eugene
AMERICA: CHANGE IT OR LOSE IT.edwin@cwi.nl (Edwin Blake) (03/07/91)
In article <1991Mar6.220940.8400@nas.nasa.gov> eugene@wilbur.nas.nasa.gov (Eugene N. Miya) writes: >Perspective does have uses. This is how we make maps, topographic maps, >and other maps. Is there any image more pervasive and full of more >useful information than a map? I don't think so. I think maps and perspective pictures serve very different functions. Moving, or stereo, perspective projected images are a good source of information about the world. But in terms of the title of this thread, perspective projection is viewer-centered while maps are object (or world) centered. A lot of processing is needed to turn a series of perspective images into maps. This difference was one of the reasons Plato called artists liars: the real nature of objects does not change with your point of view, reality persists, appearances change. Maps are true, pictures are lies. Computer graphics is more to do with the `lying' than the objective truth. >And look what we get from them: a little model of the world in orthogonal, >synopic view (these big words from my remote sensing classes). You can >adding the distances, figure out how long the drive tapes There are maps of Amsterdam which try to superimpose pictures of the buildings on a map of the city, this results in strange contortions to preserve the truth of the map and the conviction of the appearances. Perspective is largely thrown away and replaced by parallel projection. In the space of projected pictures it is very difficult to measure distance (no metric). Edwin Blake, edwin@cwi.nl Phone: +31 20 5924009 Centre for Mathematics and Computer Science (CWI) Department of Interactive Systems, Kruislaan 413, 1098 SJ Amsterdam, The Netherlands
honig@ics.uci.edu (David Honig) (03/11/91)
In article <1991Mar6.220940.8400@nas.nasa.gov> eugene@wilbur.nas.nasa.gov (Eugene N. Miya) writes: >On a long drive back from the Sierra, I pondered this topic more. >Perspective does have uses. This is how we make maps, topographic maps, >and other maps. Is there any image more pervasive and full of more >useful information than a map? I don't think so. In the vision literature at least, perspective projection means something entyirely different from binocular stereo. Binoc stereo is used to make maps from images taken at two different positions, either two eyes on either side of a nose or a moving 'plane. Perspective projection, as opposed to orthographic (parallel) projection, (the fisheye effect more pronounced with short focal-length lenses) doesn't give ranges to objects unless the object's true size is known. Then, the fact that under perspective projection, the apparent (projected) size is inversely proportional to its distance can yield distance. You can look at stereopairs that are generated with parallel (not perspective) projection. And you can look at perspective images monocularly. They are separate issues. ******************** On a separate issue related to Eugene's comments about the information in maps, take a look at Edward Tufte's recent books on displaying quantitative information. They explain what makes a good graphic objectively and are excellent reading. -- David A. Honig Quotas are for files, not people.