CI PA 2003 XIX 11 ' International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
466 
Figure 7: Screenshot of WebCAME embedded into a web 
page displayed in the Mozilla web browser showing an 
arch of the Roman baths at Sagalassos. 
the entire data is visible and close to the virtual camera: 
objects which are close to the viewer have to be rendered 
at high detail, while more distant objects, where the de 
tail is no longer visible, should be simplified. Since the 
viewpoint constantly changes in interactive applications, 
the mesh has to be adapted in real-time. The progressive 
meshes method (Hoppe, 1996) and some more involved 
techniques, such as (El-Sana and Chiang, 2000), belong to 
this class of algorithms. 
Second, geometry information stored in a straightforward 
way (i.e. coordinates for points and indices for connecting 
lines and surfaces) contains a large amount of redundancy. 
By exploiting coherence within the data set the storage size 
can be reduced. This principle has first been described by 
(Deering, 1995), more efficient techniques have been de 
veloped later, e.g. (Täubin and Rossignac, 1998, Alliez and 
Desbrun, 2001b). 
Third, if the simplified and compressed model is still too 
large to be displayed immediately after the user requested 
it, progressive transmission can be used to give the user 
a fast first impression and refine it progressively, as new 
data arrives. The user can start to operate on the (possibly 
low-quality) object with almost no delay. This feature is 
often integrated into simplification and compression tech 
niques, e.g. (Khodakovsky et al., 2000, Alliez and Des 
brun, 2001a). 
In our visualization system we combine all three concepts. 
As stated above, we use triangle meshes as the represen 
tation of 3D objects. However, even under this restric 
tion there are some properties that affect the choice of a 
particular method: generic meshes cannot be guaranteed 
to be two-manifold, and even simplifying a two-manifold 
mesh can create a non-manifold one. Therefore we will 
not rely on the two-manifold property (which greatly sim 
plifies mesh simplification), but rather use a technique that 
treats non-manifold meshes and topology simplifications 
in a natural way (see Section 4.2). 
4.2 Compressed Adaptive Multiresolution Encoding 
The CAME data structure (Grabner, 2002), which we have 
developed, is an extension and enhancement of the meta 
node approach (El-Sana and Chiang, 2000). It achieves a 
compression ratio of approximately 1:25 compared to the 
uncompressed meta-node data, while still providing view- 
dependent access to multiresolution triangle meshes based 
on an estimation of the screen-space error. 
The key idea in CAME is to identify mesh vertices by bit 
strings indicating the path to be taken in the simplification 
hierarchy to reach the leaf node corresponding to the ver 
tex. Topology compression is achieved by omitting redun 
dant prefixes of bit strings. Mesh connectivity is implicitly 
maintained by independently traversing the simplification 
hierarchy according to the bit string stored with each trian 
gle. For details we refer to the original publication. 
After reconstruction the original meshes are converted to 
the CAME format by a fully automatic encoder and stored 
into a database. The rendering system used to view the 3D 
models progressively reads the multiresolution mesh from 
the database and provides a view-dependent rendering. 
4.3 Internet-based Application 
The CAME technology has been integrated into a light 
weight web browser plugin called WebCAME (Grabner, 
2003b). It allows to progressively transmit compressed 
textured 3D models with view-dependent simplification over 
the Internet. Although being a general-purpose tool, the 
navigation facilities have been designed with virtual ar 
chaeology applications in mind. The requirements of ar 
chaeologists and of visitors an archaeological sites are re 
flected in the navigation: both viewpoint-centered navi 
gation to move around in the virtual world and object- 
centered navigation for careful exploration of single ob 
jects are provided. Figure 7 shows a screenshot of the 
WebCAME plugin with an arch from the Roman baths of 
Sagalassos. 
4.4 Smart Web Objects 
To increase the semantics of the models, the traditional 
multiresolution concept, which only handles pure geom 
etry, has been extended to preserve object identities and 
enable access to attribute data associated with the objects. 
The object identities are included in the CAME structure 
to support their compressed adaptive transmission. 
Application-specific properties of objects in the scene (meta- 1 
data) are typically stored at a separate place, such as in our i 
example the 3D MURALE multimedia database. In order < 
to access this data, it must be possible to associate each i 
piece of geometry in the scene with the appropriate meta 
data record. A simple way to solve this problem is to add 
a unique object identifier to every triangle in addition to 
color and texture information. This allows to perform the 
following important operations on the client side: * 
r 
i 
• Querying additional data about any part of the mul- t 
tiresolution mesh r 
s 
• Manipulating (e.g. translating or rotating) parts of the 
mesh c