CI PA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalva, Turkey 
509 
teaks are very 
ficant parts of 
and a virtual 
). The whole 
meshing and 
and software 
er artifact. 
fact 
point density 
necessary to 
nd/or regular 
)ly, however, 
ATOS soft- 
gic Studio or 
th qslim, for 
' about 50%, 
le model. Of 
e structure of 
le results the 
by most 3D 
ult is stored, 
. 4). For this 
) result. With 
:eware in its 
inge viewing 
el can easily 
way as the 
) an attempt 
>n the object. 
:complished, 
own texture 
m developed 
computes the differences between a triangle’s normal and the 
angles of the normals of neighboring triangles and uses it to 
classify the local curvature. The intensity of the triangle’s 
texture is then created depending on the class of the local curva 
ture. Since concave and convex curvatures are encountered, two 
different colors are used. As color printing was not possible for 
this paper, figure 6 cannot really show the fine results achiev 
able with this method. The curvature shaded object is still a 3D 
object and can be examined in a 3D viewer. When selected 2D 
views are generated, these can replace the hand drawn figures 
usually produced by archaeologists for publications. The out 
lines, as described in the following section (4.3) may be added. 
Fig. 5: Same object as in fig. 4, 
here shaded depending on local curvature 
(two different colors are used for convex and concave parts 
which cannot be distinguished in this gray tone printing). 
4.3 Object outlines in 2D visualizations 
Since manually capturing outlines for a 2D view is very tedious, 
an automatic procedure was developed at i3mainz. It detects all 
lines on a digital 3D model where the observation vector and 
the object normals are perpendicular. For a convex body they 
can also be called silhouette or contour. 
Fig. 6: Automatically generated contour (silhoutte) lines 
As 3D objects based upon triangular meshes are used, outlines 
are made up from all triangle edges belonging to front-facing as 
well as back-facing triangles. If only front-facing triangles are 
considered, outlines correspond to the outer edges of the tri 
angular mesh. Thus, finding contour lines consists of the 
following steps: 
• Choosing the desired 2D (parallel) projection. 
• Reading front-facing triangles. 
• Finding edges, belonging to one single triangle only. 
Since occlusions have to be considered, a visibility test for all 
(potential) outlines has to be performed, eliminating those lines 
that are hidden behind object parts closer to the observer. 
4.4 3D viewer for interactive inspection in the Internet 
Since users at distant locations may want to inspect the virtual 
collection, the data should be available in the Internet, too. For 
this purpose an interactive visualization tool was developed. 
Various alternatives were considered for the technology. Plug-in 
technologies (mostly proprietary tools) provide quick rendering 
solutions, optimized for the Internet usage. Alternatively, an 
approach based on Java3D was considered and finally chosen. 
This provided the necessary options for the integration of 
database access. In addition, the tool can be used as a stand 
alone application as well as an applet running in the Internet. 
Since further investigations for optimizing the visualization are 
required, the Java3D interface provides the appropriate means. 
Fig. 7: Java3D visualization tool for the Internet 
using a flat shading approach. 
Presently available shading algorithms, for instance Gouraud- 
Shading (Foley et. al., 1990), provide a good basis for 3D 
visualization including an intuitive depth perception. Those 
standard algorithms are supported by Java3D and also by most 
graphics hardware products. The shading calculations of those 
algorithms depend on the normal vectors of the surfaces. In the 
case of the Gouraud shading, the normal vector of the surfaces 
and the normal vectors at each point are used. This leads to 
smooth transitions between the polygons. In the case of visua 
lizing stone artifacts, this would eventually eliminate important 
edge information of the stone. This, of course, depends on the 
resolution of the data set as well. In order to avoid the 
elimination of this information, our first approach was to use a 
flat shading model instead, which uses only the normal vectors 
of the surfaces. This, of course, sharpened the edges but the 
overall appearance of the artifact deteriorated (see fig. 7, which