Low-Overlap Range Image Registration for Archaeological Applications
Luciano Silva
1*
, Olga R.P. Bellon
2*
, Kim L. Boyer
3
, Paulo F. U. Gotardo
2*
1
CPGEI, Centro Federal de Educac ¸˜ ao Tecnol´ ogica do Paran´ a, Brasil
2
Departamento de Inform ´ atica, Universidade Federal do Paran´ a, Brasil
{luciano,olga,paulo}@inf.ufpr.br
3
Department of Electrical Engineering, The Ohio State University, USA
kim@ee.eng.ohio-state.edu
Abstract
In digital Archaeology, the 3D modeling of physical objects
from range views is an important issue. Generally, the ap-
plications demand a great number of views to create a pre-
cise 3D model through a registration process. Most range
image registration techniques are based on variants of the
ICP (Iterative Closest Point) algorithm. The ICP algorithm
has two main drawbacks: the possibility of convergence to
a local minimum, and the need to prealign the images. Ge-
netic Algorithms (GAs) were recently applied to range im-
age registration providing good convergence results without
the constraints observed in the ICP approaches. To improve
range image registration, we explore the use of GAs and de-
velop a novel approach that combines a GA with hillclimb-
ing heuristics (GH). The experimental results show that our
method is effective in aligning low overlap views and yield
more accurate registration results than either ICP or stan-
dard GA approaches. Our method is highly advantageous
in archaeological applications, where it is necessary to re-
duce the number of views to be aligned because data acqui-
sition is expensive and also to minimize error accumulation
in the 3D model. We also present a new measure of surface
interpenetration with which to evaluate the registration and
prove its utility with experimental results.
1. Introduction
Applications for research in computer vision have expanded
significantly in recent years. Several research areas, in-
cluding Medical Imaging, Robotic Vision, and Archaeology
have presented new challenges for image registration [1].
The 3D modeling of physical objects is an important and
fundamental technique to support research in Archaeology
such as the digital preservation of cultural heritage ob-
jects [2], scanning of large statues [3], and more. Most of
these projects are focused on developing techniques to con-
*
The authors are grateful to CAPES for the financial support.
struct precise 3D object models. One of the main objectives
of the digital Archaeology is to preserve as much informa-
tion as possible about historic objects before degradation
or damage caused by environmental factors, erosion, fire,
flood, or human development. Also, some collaborative ef-
forts have supported the repair and restoration of historic
buildings, construction of virtual museums, and the analy-
sis of complex structures [4].
Typically, a 3D model is built by the registration and in-
tegration of multiple range views of an object [5, 6]. To
generate an accurate 3D model, it is necessary a robust reg-
istration method to avoid incorrect alignments and model
distortion [7]. In this process, it is essential to have a reli-
able registration method, which consists of finding the best
geometric transformation that, when applied to one view
aligns it with the other in a common coordinate system.
The most used methods for range image registration are
variations on the well-known Iterative Closest Point (ICP)
algorithm [8]. ICP is an iterative procedure minimizing the
mean squared error (MSE), computed by the sum of the
squared distances between points in one view and the clos-
est points to them in the other view. In each ICP iteration,
the best geometric transformation that aligns the two images
is calculated. The proper convergence of ICP is guaranteed
only if one of the data sets is a subset of the other; otherwise
erroneous alignments can result. Another drawback of ICP
is that it requires a good prealignment of the views in order
to converge to the best global solution. Many variants of
ICP have been proposed [5, 9, 10] to overcome these limita-
tions. In addition, comparative studies of ICP variants have
been made [10, 11], but it is difficult to evaluate these com-
parisons because there is neither a common image database
nor well defined metrics.
In archaeological applications, the registration process
is a particularly difficult task when there are many views
to be aligned and it is necessary to guarantee a precise 3D
model [1]. It is important to minimize the number of views
to be aligned, by reducing the overlap area within them.
1
1063-6919/03 $17.00 © 2003 IEEE