Three points to a perfect image

Academic collaborations are proving fruitful for the software designer Kai Krause, whose graphics tools have inspired a generation of digital artists and designers.

His company Metatools, now merged with Fractal Design and renamed Metacreations, has just published Life in the Universe, a CD-Rom on which the Cambridge physicist Stephen Hawking speculates on the origin of life on Earth and elsewhere. A team at the University of Santa Barbara created the multimedia CD.

Krause can also congratulate himself on having snapped up the Princeton University spin-off Real Time Geometry, founded by the Russian physicist Alexander "Sasha" Migdal. RTG's 3D modelling and rendering technology has potential uses range from fast-moving and complex virtual worlds on the Web to cheap 3D cameras.

To model complex surfaces, Professor Migdal and his colleagues use an economical method called Delaunay triangulation, which he has also used to crack problems in quantum physics. The method approximates any 3D shape with a mesh of carefully-chosen triangles.

Real Time Geometry has fast algorithms for removing triangles when less detail is required, or adding them back for a sharper image. Think of a packed stadium, Krause suggests. You would like to zoom into a single face and see the nose perfectly modelled with thousands of polygons. But in a wide shot showing 100,000 faces, a computer would take forever to render a detailed model of each head - and to no purpose since each face is just a tiny smudge.

Graphics programmers use tricks such as modelling objects at two or more levels of detail and switching between the models. But the RTG technology will compute a model of any required detail, on the fly. It can even interpolate new triangles to keep surfaces looking smooth at outrageous magnifications.

The technology could also be useful in virtual Web worlds, where most people's Internet connections are not fast enough to download million-polygon scenes in a fraction of a second. First you would see the main masses and surfaces in a 3D scene.The finer details,such as leaves on trees, would be filled in soon after. Without waiting for that you could move through the scene and observe from different viewpoints. It would be the 3D equivalent of the "progressive resolution" technique Web sites use for 2D images.

With a subtle twist which substitutes colour space for geometric space, RTG has applied the same mathematics to the problem of rendering and, in particular, the mapping of textures onto surfaces. Farewell, if this works, to the messy business of fitting rectangular patches of texture onto unruly freeform objects.

If one day this technology pops up in your camera, your television and your domestic robot's vision system, don't say Kai Krause didn't warn you. "We are going to license this to be inside chips, to be inside a lot of things suddenly," he says.

The company's first hardware device is a laser scanner which the RTG web pages (www.metatools.com) describe as the first practical example of true 3D photography. Krause claims it is smaller and faster than existing 3D scanners and could be mass produced for about $600: "Click, the full 3D geometry of anything you see. It certainly drove us nuts."