Fast communication
Non-ideal class non-point light source quotient image
for face relighting
$
Xiaohua Xie
a,c,d
, Jianhuang Lai
b,c,
n
, Ching Y. Suen
d
, Wei-Shi Zheng
e,c
a
School of Mathematics and Computational Science, Sun Yat-sen University, China
b
School of Information Science and Technology, Sun Yat-sen University, China
c
Guangdong Province Key Laboratory of Information Security, China
d
Centre for Pattern Recognition and Machine Intelligence,Concordia University, Canada
e
Department of Computer Science, Queen Mary University of London, UK
article info
Article history:
Received 20 January 2010
Received in revised form
9 August 2010
Accepted 12 August 2010
Keywords:
Quotient image
Face relighting
Wavelet
abstract
Quotient Image (QI) algorithm has been widely used in face recognition and re-
rendering under varying illumination conditions. One of the inaccuracies of QI
algorithm is the assumption of ‘‘Ideal Class’’, that all faces have the same surface
normal (3D shape). However, in practice this assumption is often not true. To reduce the
inaccuracy, the Non-Ideal Class Non-Point Light source QI (NIC-NPL-QI), which ignores
the ‘‘Ideal Class’’ assumption, is developed in this paper for face relighting. Unlike
that in the basic QI algorithm a fixed reference object for all test objects is used, in the
NIC-NPL-QI algorithm a special reference object for each test object is constructed, so
that the test and reference objects have similar illumination images, achieving the equal
effect of ‘‘Ideal Class’’ assumption. In the proposed method, the wavelet algorithm is
introduced to estimate an illumination image. Furthermore, the proposed NIC-NPL-QI
algorithm can handle the harmonic light and shadows. Experiments on Extended Yale B
and CMU-PIE databases show that NIC-NLP-QI algorithm obtains better quality in
synthesizing face images as compared with state-of-the-art algorithms.
& 2010 Elsevier B.V. All rights reserved.
1. Introduction
Quotient Image (QI) [3], which is defined as the rate of
albedo between a test object and a reference object, is
thought to be illumination-free. QI algorithm is practical
for single image-based face recognition and relighting.
The basic QI algorithm works under some assumptions:
(1) face is illuminated by point light source; (2) ‘‘Ideal
Class’’, all the faces have the same surface normal
(3D shape); and (3) faces are treated as Lambertian
surface without shadow. In practice, the above assump-
tions are often not true, and they may limit the appli-
cation of QI algorithm.
Recently, QI algorithm has been improved in some
ways. For face recognition, Self Quotient Image (SQI) [4],
Total Variation Quotient Image (TVQI) [10,2], and the
improved Retinex method [1] have extended QI to
illumination normalization without no-shadow assump-
tion. These algorithms treat the ratio of a face image
against its smoothed version as illumination-invariant
feature, and have attained remarkable performance for
variable lighting face recognition, but they cannot be used
for face relighting.
In this study, we focus on the improvement of QI for
face relighting, which is used to re-render a face image
Contents lists available at ScienceDirect
journal homepage: www.elsevier.com/locate/sigpro
Signal Process ing
0165-1684/$ - see front matter & 2010 Elsevier B.V. All rights reserved.
doi:10.1016/j.sigpro.2010.08.007
$
This project was supported by NSFC (U0835005, 60633030) and 973
Program (2006CB303104) in China, and NSERC (N00009) in Canada.
n
Corresponding author at: Sun Yat-sen University, School of
Information Science and Technology, Guangzhou, China.
Tel./fax: +86 20 84110175.
E-mail addresses: sysuxiexh@gmail.com (X. Xie),
stsljh@mail.sysu.edu.cn (J. Lai), suen@cenparmi.concordia.ca (C.Y. Suen),
wszheng@ieee.org (W.-S. Zheng).
Signal Processing ] (]]]]) ]]]–]]]
Please cite this article as: X. Xie, et al., Non-ideal class non-point light source quotient image for face relighting, Signal
Process. (2010), doi:10.1016/j.sigpro.2010.08.007