CIPA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
angles and distances. Besides, these representations of historical 
artifacts are enriched by online explanations, animations, music, 
video and narrations. This allows virtual museum visitors use 
this computer generated environment interactively for cultural 
research or educational purposes. In virtual museums there 
might be large numbers of dynamic virtual objects; therefore 
virtual objects should be modeled effectively. In order to 
increase realism of the virtual world, objects can be modeled 
from their images. 
In this paper we use voxel-based methods to recover the shape 
of the cultural heritage artifacts. Voxel methods consume large 
amounts of memory, for example 512 3 bytes (128 mbytes) for a 
medium size cube (512 units in each direction). Since there are 
rapid advances in hardware however, this problem is becoming 
less important and volumetric representations is becoming more 
attractive. 
The model of the 3D object can be easily acquired by shape 
from silhouettes methods in which the shape of the objects is 
recovered by intersecting the volumes. The intersection of 
silhouette cones from multiple images gives a good estimation 
of the true model. This approximate model is called the visual 
hull (Laurentini, 1995), (Matusik et al, 2000). Shape from 
silhouette methods is fast and robust; however the concavities 
and the critical areas on an object cannot be recovered with this 
method because the viewing region doesn’t completely 
surround the object. The later work to recover the shape of the 
objects from multiple images is concentrated on voxel coloring 
algorithms (Seitz and Dyer, 1997), (Culbertson et al, 1999), 
(Kuzu and Sinram, 2002). These algorithms use color 
consistency to distinguish surface points from the other points 
in the scene. They use the fact that surface points in a scene 
project into consistent (similar) colors in the input images. 
In this paper we describe several tools to refine the object’s 
visual hull. 
The organization of the chapters is as follows: In chapter 2 the 
image acquisition setup is described. The image orientation 
process is also explained. The reconstruction of the model using 
shape from silhouette technique will be described in chapter 3, 
where the refinement tools are introduced as well. Also there, 
an effective method to recover visibility information will be 
introduced. Furthermore, this information will be enhanced with 
a quality measure, by using the surface normal vector of a voxel 
in combination with the viewing direction of the image. In 
chapter 4 the refinement algorithm is explained. 
2. IMAGE ACQUISITION SETUP 
The system requirements of our experiment is simple, we use a 
CCD video-camera in order to acquire still images of the 
artifacts. Moreover, we use a calibration object to compute the 
interior orientation parameters of the camera. We place the 
object in front of a blue background. Image segmentation is a 
requirement to recover visual hull of the object. In order to 
segment object pixels from background pixels we place the 
object in front of a homogeneous blue background. We capture 
multiple views of the object resulting in a circular camera setup. 
2.1 Camera Calibration and Determination of Control 
Points 
Before acquiring the images, the camera should be calibrated. 
In our experiment we use a standard CCD video camera with 
auto focus. The focus of the camera can be fixed but we cannot 
tell whether it is unchanged since the last use. Hence, we 
calibrated the sensor using several images with a special 
calibration object which provides a good coverage of the 
objects having three perpendicular square planes and 25 control 
points on each side. 
In a second session, the object is placed inside the calibration 
frame in order to define some natural control points accurately, 
as shown in figure 1. 
A bundle block adjustment including all the images delivered 
not only the interior camera parameters, but also the coordinates 
of new points on the vase, which will serve as control points in 
the space carving processes. 
During the subsequent image acquisitions, the focus remained 
fixed. 
Figure 1: Camera calibration and control point determination 
2.2 Image Orientation 
In order to model the objects accurately, the images should be 
oriented. As mentioned above, we determined some natural 
control points on the objects surface, since we should not put 
markings on an historical artifact. They were defined in an 
arbitrarily chosen coordinate system, since there is no need to 
have the coordinates in a specific higher-level coordinate 
system. 
The images were adjusted in a bundle block adjustment process. 
We used enough tie points in all images in the circular camera 
setup to perform a bundle block adjustment, covering all 
images. Figure 2 shows an OpenGL visualization of the 
situation. We achieved very accurate results for the image 
orientations, using the previously calibrated camera. 
3. VOXEL-BASED ALGORITHMS 
3.1 Shape from Silhouette 
Shape from silhouette is a well-known approach for recovering 
the shape of the objects from their contours. This approach is 
popular in computer vision and in computer graphics due to its 
fast computation and robustness. 
As a precondition of volume intersection algorithms, the 
contour of the real object must be extracted from input images. 
In this experiment a monochromatic blue background was used