
Abstract
Faces are highly deformable objects which may
easily change their appearance over time. Not all
face areas are subject to the same variability.
Therefore decoupling the information from
independent areas of the face is of paramount
importance to improve the robustness of any face
recognition technique. This paper presents a robust
face recognition technique based on the extraction
and matching of SIFT features related to
independent face areas. Both a global and local (as
recognition from parts) matching strategy is
proposed. The local strategy is based on matching
individual salient facial SIFT features as connected
to facial landmarks such as the eyes and the mouth.
As for the global matching strategy, all SIFT features
are combined together to form a single feature. In
order to reduce the identification errors, the
Dempster-Shafer decision theory is applied to fuse
the two matching techniques. The proposed
algorithms are evaluated with the ORL and the IITK
face databases. The experimental results
demonstrate the effectiveness and potential of the
proposed face recognition techniques also in the
case of partially occluded faces or with missing
information.
1. Introduction
Face recognition is one of most challenging
research areas in machine vision and biometrics [1,
2]. The variability in the appearance of face images,
either due to intrinsic and extrinsic factors, makes the
identification problem ill-posed and difficult to
solve. Moreover, additional complexities like the
data dimensionality and the motion of face parts
causes major changes in appearance. In order to
make the problem well-posed, vision researchers
have adapted and applied an abundance of
algorithms for pattern classification, recognition and
learning. To cope for the data dimensionality, several
appearance-based techniques have been successfully
used, such as the Principal Component Analysis
(PCA) [1], Linear Discriminant Analysis (LDA) [1],
Fisher Discriminant Analysis (FDA) [1], and
Independent Component Analysis (ICA) [1]. Other
methods have been studied based on the extraction of
salient facial features by means of cascaded scale-
space filtering [3, 4, 5, 6]. Most of the times, one
missing part is the link between the features
extracted from the face images and the geometry of
the face itself.
The aim of this paper is to perform a robust and
cost effective face recognition using SIFT features
extracted from face images [7, 8, 9, 10] but also
directly related to the face geometry. In this regard,
two face-matching techniques, based on local and
global information and their fusion are proposed. In
the local matching strategy, SIFT keypoint features
are extracted from face images in the areas
corresponding to facial landmarks, such as the eyes,
nose and mouth. Facial landmarks are automatically
located by means of a standard facial landmark
detection algorithm [11, 12]. Then matching of a pair
of feature vectors is performed by a minimum
Euclidean distance metric. Matching scores produced
from each pair of salient features are fused together
using the sum rule [13]. In the global matching
strategy, the SIFT features extracted from the facial
landmarks are fused together by concatenation. Also
in this case, matching is performed by means of a
minimum Euclidean distance metric. The matching
scores obtained from the local and global strategies
are fused together using the Dempster-Shafer
decision theory. The proposed techniques are
evaluated with two face databases, the IITK and
ORL (formerly known as AT&T) face databases.
The paper is organized as follows. Section 2
briefly describes the SIFT features extraction. Local
and global matching strategies are discussed in
Section 3. Section 4 describes the fusion of local and
global matching using the Dempster-Shafer theory.
The experimental results are presented and discussed
in Section 5 and 6.
2. Overview of the SIFT feature
extraction
The scale invariant feature transform, called SIFT
descriptor, has been proposed by Lowe [8, 9] and
proved to be invariant to image rotation, scaling,
translation, partly illumination changes. The basic
idea of the SIFT descriptor is detecting feature points
efficiently through a staged filtering approach that
identifies stable points in the scale-space. Local
Face Recognition by Fusion of Local and Global Matching Scores using DS
Theory: An Evaluation with Uni-classifier and Multi-classifier Paradigm
Dakshina R. Kisku
Dr. B. C. Engineering
College, India
drkisku@ieee.org
Massimo Tistarelli
University of Sassari,
Alghero (SS), Italy
tista@uniss.it
Jamuna Kanta Sing
Jadavpur University,
Kolkata, India
jksing@ieee.org
Phalguni Gupta
I.I.T Kanpur, Kanpur,
India
pg@cse.iitk.ac.in
60978-1-4244-3993-5/09/$25.00 ©2009 IEEE
Authorized licensed use limited to: Jiangnan University. Downloaded on January 18, 2010 at 02:26 from IEEE Xplore. Restrictions apply.