bul 2004
ters on a
sation is
the left-
ientation
veen the
e Stereo
with the
stina in
theatre.
yte code
The
nders a
rom the
rameters
left-eye
ight-eye
nisation
spective
arbitrary
hus, it is
| simply
yrojector
le scene;
's with a
wd with
'ojectors
lumens
ors have
intly the
ignment
lisplays.
iey have
elativity
h means
iis is an
iginal.
ne crypt
t has an
yzantine
painted
1ensions
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004
An Italo-Canadian team from SIBA, University of Lecce and
Visual Information Technology, National Research Council
virtualised the site with a range laser scanner and
photogrammetric techniques [4]. The crypt was scanned with a
MENSI SOISIC?M range laser scanner. The spatial sampling
was 5 mm on the surface of the walls and the depth uncertainty
was 0.8 mm. The crypt was scanned in various sections of
about 2.5 m wide each. Texture information was acquired
separately with a high-resolution digital camera Nikon DIx at a
resolution of 3008 x 1960 pixels. The sections were aligned
together with the ICP algorithm and a global alignment was
performed on the complete model. These operations were
carried out with a commercial software package named
PolyworksTM. The crypt has been visualised with our system.
Figure 1 shows a column of the crypt as visualised with our
system.
Figure 2. Visualisation of the Torre dei Mesi in Trento, Italy
with the virtual theatre. Partial view of the Summer
Cycle.
The “Palace of the Good Council” or Castello del Buon
Consiglio was the symbol of the temporal power of the Prince
Bishop of Trento; now part of Italy. The actual palace consists
of many sections from the medieval and renaissance periods:
one of the best known is the “Tower of the Eagle” or Torre
dell’Aquila. The Torre dell'Aquila contains a magnificent cycle
of frescoes known as Dei Mesi i.e. “of the Months” painted by
an anonymous Bohemian painter in the 15" century. They
depict medieval life, month by month, comparing the richness
and splendour of the court with the simple life in the country.
The interior of the tower was virtualised with photogrammetric
techniques. Pictures were acquired at high-resolution by an
Italo-Canadian team from IRST — ITC and VIT - NRC with a
digital camera and a model was created with photogrammetric
techniques with a commercial software package called
ShapeCapture™ [6]. The tower has been visualised with our
system. Figure 2 shows a section of a wall corresponding to the
~
so-called “Summer Cycle” while Figure 3 shows details of the
ceiling.
Both sites are currently visualised in Italy and Canada with our
system: one virtual theatre has been deployed at SIBA (Lecce,
Italy), one at IRST — ITC (Trento, Italy) and the other one at
VIT in our laboratory. A fourth system is currently dedicated to
the mobile visualisation of industrial design at INDACO, the
School of Industrial Design of the Politecnico di Milano
(Milan, Italy).
Figure 3. Visualisation of the Torre dei Mesi in Trento, Italy
with the virtual theatre. Details of the ceiling.
3 EXPLORATION OF VIRTUAL COLLECTIONS
3.1 Content-based retrieval of images
This section presents a new algorithm for the indexation and
retrieval of images. Pictures and images are of the outmost
importance in virtual collections. They are (and will remain in
a foreseeable future) the easiest, fastest and most economical
mean for creating virtual collections. Furthermore, most three-
dimensional models are covered with textures. The textures
constitute an important visual descriptor for the model under
consideration and convey essential historical, artistic and
archaeological information. For instance, the most important
information about the Crypt of Santa Cristina described above
comes from the frescoes, which correspond to the textures. This
phenomenon is even more evident in the case of la Torre
dell'Aquila.
Images are difficult to describe. They convey a large amount of
complex and ambiguous information. The ambiguity is due to
the fact that an image is a bidimensional projection of the three-
dimensional world and by the fact that the illumination of this
world is arbitrary and cannot be controlled. Because of this
ambiguity and complexity, it is difficult to segment images and
to understand them [7]. For the above-mentioned reasons, we
propose a statistical approach in which the overall composition
of the image is described in an abstract manner.
We now depict our algorithm. The colour distribution of each
images is describes in terms of hue and saturation. This colour
space imitates many characteristics of the human visual system.
The hue corresponds to our intuition of colour e.g. red, green or
blue while saturation corresponds to the colour strength e.g.
light red or deep red.
Then, a set of points is sampled from the image. À quasi-
generates the points. In the present
t
random sequence
