TSINGHUA SCIENCE AND TECHNOLOGY
I S S N l l 1 0 07 - 0 2 14 l l 0 8 / 1 2 l l p p 6 2-6 7
Volume 18, Number 1, February 2013
Sparse Representation for Face Recognition Based on Constraint
Sampling and Face Alignment
Jing Wang, Guangda Su
, Ying Xiong, Jiansheng Chen, Yan Shang, Jiongxin Liu, and Xiaolong Ren
Abstract: Sparse Representation based Classification (SRC) has emerged as a new paradigm for solving
recognition problems. This paper presents a constraint sampling feature extraction method that improves the SRC
recognition rate. The method combines texture and shape features to significantly improve the recognition rate.
Tests show that the combined constraint sampling and facial alignment achieves very high recognition accuracy on
both the AR face database (99.52%) and the CAS-PEAL face database (99.54%).
Key words: classification; face recognition; feature extraction; face alignment
1 Introduction
Face recognition has been one of the most challenging
research areas in the world. Many face recognition
methods have been proposed, such as Principal
Components Analysis (PCA)
[1]
, Independent
Components Analysis (ICA)
[2]
, the Hausdorff
distance
[3]
, Elastic Graph-Matching (EGM)
[4]
,
and Support Vector Machines (SVM)
[5]
. These
methods achieve satisfactory results in well-controlled
environments, but their accuracy seriously degrades in
uncontrolled or moderately controlled environments,
especially when the facial pose and illumination of the
input image deviate too much from those of the training
images
[6]
. Recently, the Sparse Representation based
Classification (SRC) has attracted much attention due
to its effectiveness for face recognition with significant
Jing Wang, Guangda Su, Jiansheng Chen, Yan Shang,
and Xiaolong Ren are with the Department of Electronic
Engineering, Tsinghua University, Beijing 100084,
China. E-mail: jwang08@mails.tsinghua.edu.cn; susu@
tsinghua.edu.cn.
Ying Xiong is with the Department of Engineering and Applied
Sciences, Harvard University, Cambridge, MA 02138, USA. E-
mail: yxiong@seas.harvard.edu.
Jiongxin Liu is with the Department of Computer Science,
Columbia University, New York, NY 10027, USA. E-mail:
liujiongxin@gmail.com.
To whom correspondence should be addressed.
Manuscript received: 2012-07-16; accepted: 2012-12-05
illumination and expression variations
[7]
. This method
uses the theory of compressive sampling
[8]
to exploit
the discriminative nature of the sparse representation
for classification. With proper selection of the training
samples and the number of features, the SRC algorithm
achieves good recognition results even with serious
variations in the illumination conditions or expressions.
The SRC algorithm has given promising recognition
results on public face databases including the extended
Yale face database B
[9]
and the AR face database
[10]
.
Wright et al.
[7]
claimed that if the recognition
sparsity was properly used, the choice of features
would no longer be critical. They then proposed a less-
structured feature called Randomfaces and validated
through experiments that this could achieve similar
results as conventional features. However, different
parts of a face contain different amounts of information.
If the features are selected for high-informational
areas, the recognition efficiency will be improved
for the same feature dimension. Thus, this paper
presents a feature extraction method called constraint
sampling that obtains a fixedface feature value using key
facial points. Comparison with results using traditional
features demonstrates that a proper uneven selection
of features will capture the main characteristics
and achieve better recognition performance. Another
advantage of constraint sampling is that the face
images are better aligned through the key point locating
process, so it efficiently overcomes the alignment