Web Aware Graphics 

by Dieter Fellner (TU Braunschweig) 

Evolving from simple collections of text documents to multimedia databases,
current _Digital Libraries_ (DLs) have to cope with a broad range of
different media formats. Among more traditional media like text, audio and
video, 3D models are becoming increasingly important in many practical
applications thereby raising a number of fundamental problems.

The complexity of a DL containing a large number of inter-linked digital
documents can only be managed if efficient _markup_ and _retrieval_
techniques are employed. But with the availability of 3D scanning devices,
and in the lack of a satisfying standard for the high-level description of
3D surface data, textured polygonal models have become the de-facto standard
for 3D models. Undoubtedly representing a very important class of 3D models, 
they unfortunately have certain drawbacks. Polygonal models are ``flat'', 
because structural information about the model cannot be expressed in terms of
polygonal faces. On a syntactic level, there's no difference between the
model of a spoon, for example, and the model of a chair. Yet the way how
these objects are constructed is very different.

Another problem with polygonal models is that curved surfaces can only be
represented as tesselated approximations, and file sizes grow quickly with
quality requirements. Finally, the retrieval of 3D data from a DL is
performance critical if large datasets are transmitted over low-bandwidth
networks and have to be displayed at interactive frame rates. Therefore,
efficient techniques for {\em spatial retrieval_ are to be developed, that
permit to quickly determine the portion of a huge scenery that is contained
in the current 'region of interest'. Only this way a guaranteed interaction
frame rate can be maintained.

The talk will briefly cover current techniques to cope with large polygonal
meshes in a distributed environment before presenting an alternative to the
triangle-based modeling and exchange of geometric information. The presented
novel approach based on generative modeling is superior to current solutions
with respect to the utilization of network bandwidth and interactive
navigation in large geometric models.