CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
2 DATA PROCESSING 
In earlier work (Adler et al., 2001) a scanning technology 
was described, which directly acquired the profile line by 
projecting a laser-line on the sherd. This laserline was cap 
tured by a camera for the inner side and a second cam 
era for the outer side. The merged images of these two 
cameras contained the profile line. This system had draw 
backs concerning its portability, ease of use and automa 
tion. Therefore we choose the Eyetronics Shape Snatcher 
Technology (Cosmas et al., 2001) (see Figure 3), because 
it is portable and can be operated without expert knowl 
edge. It consists of a CCD-camera and a flashlight to ac 
quire the shape based on the structured light principle. The 
image, together with the knowledge about the pattern and 
its relative position to the camera, are used to calculate the 
coordinates of points belonging to the surface of the ob 
ject (Kampel and Sablatnig, 1999). Since the 3D-scanner 
can only capture one side of the sherd per scan, the inner 
side and the outer side have to be scanned separately to 
reconstruct a complete 3D-model of the sherd. 
Figure 3: Eyetronic’s ShapeCam consisting of a CCD- 
camera (left) and a flashlight mounted on the top-right part 
of the handling frame. 
2.1 3D Data - VRML 
The 3D-data acquired by the 3D-scanner is stored as 3D- 
surface, which consists of 3D-points (vertices) that are con 
nected in form of triangles (called faces). The 3D-model 
contains the color information (called texture) for each face 
for visualization. These vertices and faces are stored in an 
indexed list (E). 
There are different types of file formats for storing 3D-data 
(e.g. AutoCAD, Wavefront, Openlnventor). We have cho 
sen VRML (Nadeau, 1999), which is software independent 
and can be viewed with a web-browser with a VRML-plugin 
that is free of charge. 
The pottery data set we use for experiments consists out of 
the recorded objects described in table 1. The number of 
pieces per view is 2, because for every piece an inner view 
and an outer view has been acquired. 
For storage in a database we use polygonal geometrical 
Box 1 
Box 2 
Box 3 
Number of pieces 
21 
12 
26 
Number of views/piece 
2 
2 
2 
Vertices/View 
4.000-9.000 
3.000-8.000 
Table 1 : MURALE Pottery dataset 
objects (Nadeau, 1999). These geometrical objects are de 
scribed by an indexed list C v of n vertices p, = (Xi,yi,Zi) T 
Xi ,yi and Zi are the coordinates of the vertices of the sur 
face of the object in centimeter. 
f-'v Pn} (1) 
The vertices of C v are connected by faces /, which are 
stored as list Ef. 
£/ = {fi, ■ ■ ■, /m}, fk — {ki,k 2 , k 3 ) (2) 
Optionally the normal vectors C n — {ni,..., n n } and the 
color information C c = {ci,..., c n }, c* = (red,green, blue) 
is stored, which is not necessary for further calculation, but 
it is used for visualization purposes. The normal vectors 
E n are used for the estimation of the rotational axis. If 
they are not provided by the scanning software, they are 
estimated by using the triangulated data. 
So the 3D-model of the view is described by an indexed 
list of vertices, faces, normal vectors and color informa 
tion. sfieTcCyieuj {Ey, E f, Eji, Efi) . 
To double the performance of processing the data, we use 
two different types of coordinate systems. For translation 
and rotation we use the cartesian coordinate system, be 
cause the translation is done by an addition. To estimate 
the two angles (azimuth and elevation), which is required 
for rotation we use a spherical coordinate system. Points 
p and vectors v in R3 using the spheric representation are 
described by the azimuth 9, elevation 0 and the distance r 
to the origin point in the cartesian coordinate system. This 
representation has been chosen for the rotation, because ro 
tation can be done by an addition. Cartesian p-(ai, y, z) T 
and spheric: p=($, 0, r) T notation. 
2.2 Profile line 
The rotational axis rot leads to the exact position of a frag 
ment on the original vessel. Therefore the rotational axis is 
required for the estimation of the profile line and for regis 
tration (Sablatnig and Kampel, 2002) of the inner and outer 
view. 
The profile line, which is used by archaeologists for clas 
sification and reconstruction, is defined below: 
• A profile line (profile) is the cross-section of the 3D- 
model of the sherd {sherd) and an intersecting plane 
ej. This intersecting plane e l , is defined by rot and the 
direction i, so that e* intersects the sherd. 
• The intersection at an index i maa: , where the sherd has 
the maximum height h ma x = max{hi) is the profile 
line with the longest arc length and is called longest 
profile line (profile ). The index specifies the direc 
tion. 
• The height hi is defined as distance between two points 
of the surface of the sherd parallel to the rotational 
axis rot. The index i, where the height has its maxi 
mum is called h rnax . 
A sherd, its rotational axis and the estimated longest profile 
line is shown in Figure 4.