COL 9(5), 051002(2011) CHINESE OPTICS LETTERS May 10, 2011
Parallel sub-neural network system for hand vein pattern
recognition
Xue Yuan (袁袁袁 雪雪雪), Yongduan Song (宋宋宋永永永端端端)
∗
, and Xueye Wei (魏魏魏学学学业业业)
Center for Intelligent Systems and Renewable Energy, Beijing Jiaotong University, Beijing 100044, China
∗
Corresp onding author: ydsong@bjtu.edu.cn
Received October 9, 2010; accepted November 19, 2010; posted online April 22, 2011
A hand vein authentication system in which the identity of an individual can be readily confirmed upon
gripping a handle is proposed. This recognition method incorporates infrared light-emitting diode (LED)
onto a door handle and sets a charge-coupled device (CCD) camera on the other side of the hand. It
builds on fuzzy c-means clustering and parallel neural networks (NNs); moreover, it is expected to solve
the pattern recognition problem in large-scale databases using NNs due to its self-learning and parallel
pro cessing capabilities and by effectively incorporating training patterns. The experimental results validate
the efficiency of the proposed algorithm.
OCIS codes: 100.0100, 110.0110, 150.0150.
doi: 10.3788/COL201109.051002.
Vein authentication technology is a new biometric tool
that has recently attracted considerable attention
[1−9]
.
The pattern of veins is generally known to be hardwired
into the body at birth and remains relatively unaffected
by aging, except for predictable growth, similar to that
of fingerprints.
Vein geometry is based on the fact that an individ-
ual’s vein pattern is distinct. The deoxidized hemoglobin
in veins absorbs light at a wavelength of approximately
7.6 × 10
−4
mm within the near-infrared (NIR) area.
When the infrared ray image is captured, only the blood
vessel pattern containing the deoxidized hemoglobin is
visible as a series of dark lines. Based on this feature,
a vein authentication device translates the black lines of
the infrared ray image as the blood vessel pattern of the
hand and then matches it with the previously registered
blood vessel pattern of the individual.
Kumar et al.
[1]
thoroughly discussed the use of hand
vein triangulation and knuckle shape for vein recogni-
tion. Wang et al.
[2]
presented a multi-resolution wavelet
algorithm for hand vein pattern recognition. Zeng et
al.
[3]
developed a method using curvelet transform to
extract features from vein patterns as well as principal
component analysis (PCA) and nearest-neighb or classi-
fier method for vein recognition. In our previous work
[4]
,
we established a two-step minutia-based method for vein
recognition. All these proposed methods demonstrated
good experimental performance. However, problems in
the vein recognition field remain. Such problems are as
follows: (1) The current methods are not user-friendly
as users must place their hands on the identification
device and wait for authentication. (2) Most existing
systems are unable to recognize a pattern from a very
large database using neural networks (NNs) effectively,
especially when there are numerous patterns registered
in the database, indicating the significance of large-scale
NNs. (3) Systems have yet to develop the capacity to in-
corporate and adjust training patterns in large databases
and in a timely manner.
A new grip-type hand vein authentication system that
is more convenient to use than previously developed
systems is herein proposed. The incorporation of this
authentication system onto a door handle allows the
identity of an individual to be readily confirmed upon
gripping the handle, without any additional maneuvers,
such as placing the hand on the identification device. The
proposed system differs from other hand vein recognition
systems in that an infrared light is transmitted from the
back of the hand. The hand is placed between the in-
frared light source and a camera. As hemoglobin in the
blood absorbs the infrared light, the pattern of veins at
the back of the hand is captured as a pattern of shadows.
This new recognition method is based on fuzzy c-means
(FCM) and parallel NNs, which are established as fol-
lows: (1) Based on the locations of extracted minutiae
points, the system can classify the registers into several
clusters using the FCM method. (2) A small-scale three-
layer back-propagation NN (BPNN), called a sub-NN, is
then created in each cluster, and the process of learning
and recognition in each sub-NN is conducted indepen-
dently and synchronously. The system can effectively
recognize a pattern from very large databases because
of the algorithm’s self-learning properties and parallel
processing capability. Furthermore, the system can in-
corporate training patterns, thus avoiding the exp ensive
computational cost of retraining the whole system.
An example of the original images in the database is
shown in Fig. 1. The system consists of one infrared
light-emitting diode (LED) array and a charge-coupled
device (CCD) camera. A band-pass filter was set be-
tween the infrared LED and the CCD camera to avoid
the effects of varied illumination conditions.
This study focused on recognizing the hand vein pat-
tern at the back of the hand because palm veins are com-
pressed and deformed in the gripping action, rendering
Fig. 1. Hand vein images.
1671-7694/2011/051002(4) 051002-1
c
° 2011 Chinese Optics Letters