A novel face recognition method: Using random weight networks
and quasi-singular value decomposition
Wanggen Wan
a
, Zhenghua Zhou
a
, Jianwei Zhao
b
, Feilong Cao
b,
n
a
School of Communication and Information Engineering, Shanghai University, Shanghai 200072, PR China
b
Department of Mathematics, China Jiliang University, Hangzhou 310018, Zhejiang Province, PR China
article info
Article history:
Received 16 June 2013
Received in revised form
19 June 2014
Accepted 26 June 2014
Available online 18 October 2014
Keywords:
Face recognition
Feature extraction
Quasi-singular value decomposition
Random weight networks
abstract
This paper designs a novel human face recognition method, which is mainly based on a new feature
extraction method and an efficient classifier – random weight network (RWN). Its innovation of the
feature extraction is embodied in the good fusion of the geometric features and algebraic features of the
original image. Here the geometric features are acquired by means of fast discrete curvelet transform
(FDCT) and 2-dimensional principal component analysis (2DPCA), while the algebraic features are
extracted by a proposed quasi-singular value decomposition (Q-SVD) method that can embody the
relations of each image under a unified framework. Subsequently, the efficient RWN is applied to classify
these fused features to further improve the recognition rate and the recognition speed. Some
comparison experiments are carried out on six famous face databases between our proposed method
and some other state-of-the-art methods. The experimental results show that the proposed method has
an outstanding superiority in the aspects of separability, recognition rate and training time.
& 2014 Elsevier B.V. All rights reserved.
1. Introduction
F ace r ecognition has attracted m uch att ention in recent y ears due
to its inexpensi ve, convenient and hassle-free advantages. It has
widespread applications, such as smart cards, telecommunication,
database security, medical records, and digital libraries [27 ].There
are two key steps in the face recognition: feature e xtraction and
classification. The aim of feature extr action is mainly to give an
effective representation of each image, which can r educe the compu-
tational complexity of the classification algorithm and enhance the
separability of t he images to get a higher recognition r ate. While the
aim of classification is to distinguish those extracted features with a
good classi fier . Therefore, an effective face recognition system greatly
depends on the appropriate representation of human face features and
the g ood design of classifier .
It is well known that image features are usually classified into
four classes: Statistical-pixel features, visual features, algebraic fea-
tures, and geometric features (e.g. tr ansform-coefficient features),
where the latter two are often applied in face recognition system. Up
to now , many geometric-feature based methods hav e been proposed
to acquir e a higher level of separability through transforming the
space-domain of the original image into another domain. At the early
stage, wavelet transform is popular and widely applied in face
recognition system for its multi-resolution character, such as
2-dimensional discrete wavelet transform [6], discrete wavelet trans-
form [8], fast beta wa velet networks [1 2], and wave let based featur e
selection [9,2 4,26,29]. Although wa v elet transform is suitable for
detecting singularities in the image, it fails to represent curved
discontinuities. Fortunatel y, Donoho and Duncan have proposed
another transform-based method, called curv elet transform, t o
improv e the directional capability [4]. That is, it can represent edges
with singularities well and increase anisotrop y with decreasing scale
as well. Therefore, curvelet transform has been widely used in face
recognition, such as curvelet based moment method [14], curvelet
based face recognition [1 5,16], and curvelet based image fusion [21].
On the other hand, the algebraic features of images can r eflect the
intrinsic properties of images stably . Therefore, they have been
considered as valid features for face recognition [10].Asoneofthe
effective algebraic-feature based methods, singular value decompo-
sition (SVD) method [7] was applied in face recognition to extract
feature vectors [3,23]. It can represent algebraic features from space-
domain w ell. However, images usually come from the same class in
practical applications, so they are often related to each other . Hence it
is no t enough t o just take SVD method in images as SVD method is
appliedoneachimageseparately.
As we all kno w , rela ted liter atur es only use one of the ab ov e
features in the process of extracting face features. In this paper , we will
propose a new method for extracting compound features that consist
of geometric featur es and algebr aic featur es. At first, we use FDCT to
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/neucom
Neurocomputing
http://dx.doi.org/10.1016/j.neucom.2014.06.081
0925-2312/& 2014 Elsevier B.V. All rights reserved.
n
Corresponding author. Tel.: þ 86 571 86835737.
E-mail addresses: wanwg@staff.shu.edu.cn (W. Wan),
zzhzjw2003@163.com (Z. Zhou), zhaojw@amss.ac.cn (J. Zhao),
feilongcao@gmail.com (F. Cao).
Neurocomputing 151 (2015) 1180– 1186