1
Corresponding author: cslinzhang@tongji.edu.cn
A NOVEL 3D EAR IDENTIFICATION APPROACH BASED ON SPARSE
REPRESENTATION
Zhixuan Ding, Lin Zhang*
1
, and Hongyu Li
School of Software Engineering, Tongji University, Shanghai, China
ABSTRACT
Recently, ear shape has attracted tremendous interests in
biometric research due to its richness of feature and ease of
acquisition. In this paper, we present a novel 3D ear
identification approach based on the sparse representation
framework. To this end, at first, we propose a
template-based ear detection method. By utilizing such a
method, the extracted ear regions are represented in a
common standard coordinate system determined by the
template, which facilitates the following feature extraction
and classification. For each 3D ear, a feature vector can be
generated as its representation. With respect to the ear
identification, we resort to the l
1
-minimization based sparse
representation. Experiments conducted on a benchmark
dataset corroborate the effectiveness and efficacy of the
proposed approach. The associated Matlab source code and
the evaluation results have been made online available at
http://sse.tongji.edu.cn/linzhang/ear/srcear/srcear.htm
.
Index Terms—Biometrics, 3D ear recognition, sparse
representation, Iterative Closest Point
1. INTRODUCTION
Among all the biometric identifiers, ear is a relative new
member and it has been proved viable for its desirable
properties such as universality, uniqueness and permanence
[1, 2]. Besides the traditional 2D ear recognition [2, 3, 4, 5],
there also exists a technology to acquire ear data by using a
3D sensor which provides both 2D and 3D data for an ear.
Compared with 2D data, 3D ear data contains more
information about ear shape and is not sensitive to
illumination and occlusion.
Recently, several researches for 3D ear recognition
have been conducted. In [6], Chen and Bhanu developed a
3D ear recognition system. The algorithm they suggested is
based on a 2-step Iterative Closet Points (ICP) [7]
framework and all the ear regions are extracted from profile
images manually. To make their system automatic, they
presented an ear detection algorithm by using an ear shape
model in their later work [8], in which the first coarse ICP
step is performed on two extracted ear helixes and the
second fine ICP step is further applied on two corresponding
ear point clouds by setting the result gained from last step as
initial translation. In 2007, Yan and Bowyer also proposed
an automatic ear recognition approach by applying a 3D ICP
algorithm [9]. In their work, they tried to locate the ear pit
and then used active contour algorithm [10] to extract the
ear contour. Their recognition process is no different from
any other ICP based approaches. At the same year, Chen
and Bhanu improved their work by introducing a Local
Surface Patch Representation [11]. 3D ear recognition was
also investigated by Islam and Mian [12]. For ear detection,
they adapted the Viola-Jones object detection algorithm [13]
and for feature extraction, they adopted the Local 3D
Feature (L3DF) scheme proposed in [14].
In another aspect, as an effective classification tool [15,
16], sparse representation has also been introduced to the 3D
biometrics field. For instance, in [17], Li and Jia proposed a
3D face recognition approach based on sparse representation
and promising results were reported.
From the aforementioned introduction, it can be seen
that most existing 3D ear or 3D face recognition methods
are based on ICP. While ICP is a feasible 3D matching
model for the one-to-one verification, it is not quite
appropriate for the one-to-many identification case. If there
are multiple samples for each subject in the gallery set, the
recognition based on ICP usually would have to match the
test sample to all the gallery samples. With the number of
gallery samples rising, the performance of ICP-based
methods will markedly slow down. Since the task of
recognition is essentially to find a single individual out of
the entire dataset, Wright et al. [16] proved that the
recognition based on sparse representation framework is
more suitable to solve the multiple samples case efficiently.
Li and Jia have also shown the feasibility of applying the
sparse representation framework to 3D face recognition [17].
However, to the best of knowledge, so far there is no work
reported to apply sparse representation for 3D ear
recognition.
Based on these considerations, in this paper, we
propose a novel 3D ear identification approach based on
sparse representation. Our approach consists of three
components, ear detection, feature extraction, and
classification. For ear detection, we propose a template
based scheme which is robust to ear pose change. For
feature extraction, we adopt an effective PCA-based
descriptor proposed in [20]. Our approach takes 3D point
cloud as input and no extra color image is required. The
performance of the proposed approach is examined on the
benchmark dataset and is compared with the ICP based
4166978-1-4799-2341-0/13/$31.00 ©2013 IEEE ICIP 2013