Automatic 3D Face Recognition Using Fourier Descriptors
Eyad Elyan
School of Computing,
Robert Gordon University
Aberdeen, AB251HG, UK
Email: e.elyan@rgu.ac.uk
Hassan Ugail
School of Informatics
University of Bradford
Bradford, BD71DP, UK
h.ugail@brad.ac.uk
Abstract—3D face recognition is attracting more attention
due to the recent development in 3D facial data acquisition
techniques. It is strongly believed that 3D Face recognition
systems could overcome the inherent problems of 2D face
recognition such as facial pose variation, illumination, and
variant facial expression. In this paper we present a novel
technique for 3D face recognition system using a set of
parameters representing the central region of the face. These
parameters are essentially vertical and cross sectional profiles
and are extracted automatically without any prior knowledge
or assumption about the image pose or orientation. In addition,
these profiles are stored in terms of their Fourier Coefficients
in order to minimize the size of input data. Our approach
is validated and verified against two different datasets of 3D
images covers enough systematic and pose variation. High
recognition rate was achieved.
Keywords-2D Face Recognition; 3D Face Recognition; 3D
images
I. I NTRODUCTION
Most of the research work done on the area of face
recognition was based on 2D representation of facial data,
hence a wide range of 2D algorithm are available in the
literature [1].
3D face recognition is attracting more attention in the
recent years due to two important factors. Firstly, because of
the inherent problems with 2D face recognition system that
appears to be very sensitive to facial pose variation, variant
facial expression, and lighting and illumination. For more
information about current work in the 3D recognition area
see [2], [3]. Secondly, due to the recent development in the
3D acquisition techniques such as 3D scanners, infrared, and
other technologies that makes obtaining 3D data relatively
much easier.
3D face recognition is still faced with several challenges.
One of these is the lack of an available benchmark dataset
that could be used for experimentation and thus provide a
clear and solid indication of the robustness of any recogni-
tion system. This is very clear if we knew that till 2003 the
number of persons in datasets used for 3D face recognition
experimentation didn’t reach 100 [3]. In addition, very few
papers deals with pose and expression variation [3], [4], [5].
Several other challenging could be listed regarding the
progress and development of 3D face recognition systems
such as: Feature points allocation (this is still a debatable
topic) that is also sensitive to the quality of data. Sampling
density of the facial surface, and accuracy of the depth
(e.g. no clear answer how much dense should a facial
surface be? to accurately represent an individual face) [1].
No standard testing protocol is available to compare between
different Face Recognition systems. Age factors, the size
of the database, and efficiency of used algorithms are also
considered as major challenges [3].
The rest of the paper is organized as follows: in the
following section we will briefly review methodologies
deployed in 3D Face Recognition systems. The next section,
we will discuss our 3D image processing technique and our
work on characterizing and allocating certain facial features
automatically such as the tip of the nose, symmetry profile
and cross-sectional profiles in the central region of the face.
In addition, we will discuss our matching algorithm and the
results, draw conclusion upon that and suggest future work
for further improvement.
II. PREVIOUS WORK
Face recognition may be considered as a template match-
ing with a high dimensionality. Dimensional reduction tech-
nique is often used to reduce dimensionality in order to
reduce computation cost. Kirby used Principle Component
Analysis (PCA) in addressing the problem [6]. Among many
various algorithm PCA [7] has become a corner stone in
2D face recognition system. However 2D Face Recognition
systems are unable to overcome the problems mentioned
earlier.
Several approaches are used in the literature for 3D Face
Recognition. Some of these are based on the segmentation
of the face into meaningful points, lines and regions. Others
are considered as model based approaches using information
about texture, edges, and colors. Some techniques are con-
sidered as a profile-based techniques where multiple profiles
comparisons are carried out, by which a set of profiles
are compared against each other, such profiles might be
symmetry ones, transverse, vertical or even cross-sectional
[8]–[10].
Among the existing approaches for addressing 3D face
recognition systems is the use of extended Gaussian Image
2009 International Conference on CyberWorlds
978-0-7695-3791-7/09 $26.00 © 2009 IEEE
DOI 10.1109/CW.2009.48
246