CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalva, Turkey
The smoothness of the visualization of the reconstructed
3D-model depends on n. Larger n means smoother, but
slower to display. Experiments have shown that 180 < n
< 360 is the best trade off between performance and qual
ity of the visualization of the reconstructed object.
Reconstruction is done by rotating the profiCe 360° about
the rotational axis using R z (7).
object — {projiie * R z (7i)}, i = 1.. .n,
7i = 0, 7 i+ i = 7i + A7, A7 = 2 * rx/n (5)
Figure 10a shows the reconstruction of a pot and Figure 10b
shows the reconstruction of fragment. The reconstruction
of these two Figures were based on the longest profile line
from Figure 8.
4 RESULTS
Experiments were done on the 33 sets of 3D images stored
in box 1 and 2 and 26 real data sets from box 3 of archaeo
logical fragments given by archaeologists for testing. Each
set contained one image of the inner half and one of the
outer half of the sherd. In 29% of the sherds the estimation
of the rotational axis returned a correct result. 31% of the
results had two different types of minor errors, which are
still acceptable for further processing.
The first acceptable error was a to large distance between
the inner and out half (2 to 3 cm). The second acceptable
error was a slightly twisted (less than 10°) inner half com
pared to the orientation of the outer half. These two errors
have been observed on small sherds or sherds with a small
curvature (Sablatnig and Kampel, 2002). For 7 sets the es
timation of the rotational axis did not have a correct result,
because the sherd were to small, to flat, contained a handle
or were part of a bottom fragment. All of these 7 sets have
normal vectors, which do not point at the rotational axis.
So the estimation of the rotational axis was not done cor
rectly.
The success rate for correct extraction of the profile line
and consequently the percentage of sherds, which is used
for further classification is 50% of the sherds found at the
excavation site. This has to be seen with respect to man
ual archivation done by archaeologists (Orton et al., 1993):
for coarse ware 35% (Degeest, 2000) and for fine ware
50% (Poblome, 1999) of the findings are used for further
classification. It depends on the ratio between bending of
the curvature (Matas et al., 1995, Bennett and MacDon
ald, 1975) and the fragment and its diameter (Sablatnig
and Kampel, 2002) (e.g. handle, flat fragments like bot
tom pieces, small size, etc.).
The execution time using a prototype written in MAT LAB
running on a Pentium III 1 GHz is less than a minute per
sherd. The estimation of rot takes 70% to 80% of the exe
cution time for processing one sherd described by the inner
and outer view. Comparing the execution time for the ex
traction and segmentation of profile lines to the time used
by archaeologists drawing a profile line by hand shows that
the number of classification per day can be increased dra
matically.
The estimation of the rotational axis will also be used to
reconstruct whole objects from several sherds. Figure 10a
displays a reconstructed pot (gray object) out of one frag
ment (dark object) based on the profile line (light line)
and its axis of rotation (dashed line). Figure 10b shows
a detailed part of the same object as Figure 10a. Table 2
(b)
Figure 10: (a) Reconstructed (gray) pot and (b) fragment,
cross-section (light gray line), recorded fragment (dark
gray) and its rotational axis (vertical dashed line).
shows successfully estimated features for further classifi
cation of box 1 and box 2. These features are the diameter
at the highest point of the sherd (rim-diameter rdm in cm).
The maximum diameter of thb sherd orthogonal to the ro
tational axis (wall-diameter wdm in cm). The diameter
at the lowest point of the sherd (bottom-diameter bdm in
cm). The overall height (h max in cm) of the sherd and
the characteristic ratio crat = h : rdm. Box 3 contains
Box
Nr
rdm
wdm
bdm
hmax
crat
1
04
25,16
75,61
24,94
7,12
0,33
1
08
31,78
50,16
32,08
5,63
0,63
1
16
30,68
49,32
30,9
6,68
0,62
1
17
32,64
42,31
34,32
8,81
0,77
1
18
32,34
31,06
32,1
4,68
1,04
1
19
28,9
33,80
29,38
9,15
0,85
1
20
26,78
30,15
27,32
7,09
0,89
1
22
22,8
27,62
23,86
6,04
0,83
1
23
24,58
44,61
25,32
7,4
0,55
2
01
20,68
54,22
20,8
7,05
0,38
2
02
32,58
35,02
32,9
8,94
0,93
2
04
12,54
35,66
17,38
6,9
0,35
2
05
19,92
25,79
20,56
6,15
0,77
2
06
6,84
22,85
7
10,15
0,30
2
09
24,66
19,05
25,76
6,96
1,29
Table 2: Results of proper registered and orientated sherds.
