ROBUST FEATURE ENCODING FOR AGE-INVARIANT FACE RECOGNITION
Xiaonan Hou, Shouhong Ding, Lizhuang Ma
Department of Computer Science and Engineering
Shanghai Jiao Tong University, China
xnhou1989@163.com, dingsh1987@gmail.com, ma-lz@cs.sjtu.edu.cn
ABSTRACT
Large age range is a serious obstacle for automatic face recog-
nition. Although many promising results have been reported,
it still remains a challenging problem due to significant intra-
class variations caused by the aging process. In this paper, we
mainly focus on finding an expressive age-invariant feature
such that it is robust to intra-personal variance and discrim-
inative to different subjects. To achieve this goal, we map
the original feature to a new space in which the feature is
robust to noise and large intra-personal variations caused by
aging face images. Then we further encode the mapped fea-
ture into an age-invariant representation. After mapping and
encoding, we get the robust and discriminative feature for the
specific purpose of age-invariant face recognition. To show
the effectiveness and generalizability of our method, we con-
duct experiments on two well-known public domain databas-
es for age-invariant face recognition: Cross-Age Celebrity
Dataset (CACD, the largest publicly available cross-age face
dataset) and MORPH dataset. Experiments show that our
method achieves state-of-the-art results on these two chal-
lenging datasets.
Index Terms— face recognition, age-invariant, intra-
personal robustness, feature encoding
1. INTRODUCTION
In recent years, age-invariant face recognition research be-
comes an attractive topic due to its wide range of application
scenarios. Age information is useful in many application-
s, such as age-specific human-computer interaction, securi-
ty surveillance monitoring, age-based face images retrieval,
automatic face simulation and intelligent advertisement sys-
tem etc.. Although significant progress has been made in face
recognition, age-invariant face recognition still remains a ma-
jor challenge in real world applications such as face recogni-
tion systems, in which age-invariant face image analysis is a
main obstacle. The difficulty of this problem, to a great ex-
tent, arises from the fact that the face appearance of a person
This work is supported by a joint project of Tencent and Shanghai Jiao
Tong University. It is also partially supported by the National Natural Science
Foundation of China under Grant (No.61133009 and No.61472245).
Fig. 1. Examples of CACD dataset [1]. Each column corre-
sponds to one single person’s images of year 2004, 2008 and
2013.
is subject to remarkable change caused by the aging process
over time, as shown in Figure 1.
There are lots of researches on age-related face analysis
focusing on age estimation [2, 3, 4, 5] and aging simula-
tion [6, 7, 8]. However, work on age-invariant face recog-
nition is limited. In recent years, several promising methods
have been proposed such as [1, 9, 10, 11, 12, 13]. Method
in [9] uses gradient orientation pyramids (GOPs) for feature
representation to verify faces with large age differences. [10]
combines SIFT [14] and MLBP [15] with a random sampling-
based fusion framework to improve age-invariant face recog-
nition performance, [16] is used as features and a variation
of random subspace LDA approach (RS-LDA) [17] is used
for classification. [1] proposes a probabilistic model with two
latent factors: an identity factor that is age-invariant and an
age factor affected by the aging process, then the observed
appearance can be modeled as a combination of the compo-
nents generated based on these factors. [11] proposes a nov-
el coding framework by encoding the low-level feature of a
face image with an age-invariant reference space. More re-
cently, [12] proposes a new deep Convolutional Neural Net-
work (CNN) model for age-invariant face verification, they
also introduce two tricks to overcome insufficient memory