A systems that are based on behavioral patterns

A review of nger vein biometrics authentication
System
Sara Daas
Laboratory L.E.R.I.C.A Electronics Department
University of Badji Mokhtar Annaba , Email:[email protected] Mohamed Boughazi
Laboratory L.E.R.I.C.A Electronics Department
University of Badji Mokhtar Annaba
Abstract —In nowadays, with the fast development in the eld
of electronic information technology, the identity verication and
security of data is a critical main problem. Due to this fact, the
biometric recognition has gaining popularity as it provides a high
security , reliable and robust approach for personal identica-
tion. As a new biometric technique, nger vein recognition has
attracted lots of attention from researchers. In this paper a survey
has been done on nger vein biometrics recognition taken from
existing research work on image preprocessing, feature extraction,
classication stage and the performance parameter of biometric
system the Equal Error Rate (EER).The main aim of this study
is to enlighten nger vein authentication research domain.
Keywords —Biometrics, nger vein recognition, preprocessing,
feature extraction, classication, EER .
I. IN T RO D U C T I O N
Biometrics is an emerging technique that enables us to
verify the identity of an individual by using at least one of
his or her personal characteristics 1.The biometric systems
are classied into three categories .biometric systems that are
based on behavioral patterns such as handwriting and keystroke
dynamics, that are based on physiological patterns such as
ngerprint, face and iris and the systems that are based on
hidden patterns such as the electrocardiogram (ECG) and MRI
images 2.Those systems somehow have several inconvenient;
the features with age may not be clear for identication and can
be easily stolen, lost or borrowed 3.A pattern of this type of
biometric trait is ngerprint and voice 1 ,also iris recognition
is considered as least user cause it is uneasiness to human eyes
due to the brightness of light during the biometric capture
process 4.MRI and ECG are expensive and complicated
techniques 5, 6 .The handwriting is a natural and familiar
way of conrming identity but not secure6 . TABLE.I show
more detailed comparison between some previously mentioned
biometric systems.
Recently in the literature, a biometric method based on
vascular patterns such as hand vein and peculiarly nger vein
has interested the researchers. A nger vein biometric system
came into existence after the invention of nger print. The
veins are within of the human nger. Finger vein features
offering varied other favorite advantages and these include:
The nger vein patterns are exceptional for every
person, even identical twins. So, it offers a great
distinction between each individual.
The patterns of ngers veins do not change with time,
they are permanent.
The nger vein patterns are invisible to humans eyes.
Consequently, they are not obscure and difcult to be
replicated of the fact that it is positioned underneath
the human skin.
The nger-vein patterns acquisition is considered to
be easy to use . The vein pattern images captured non
invasive. The nger vein device is contactless sensors,
so his concept ensuring hygiene and convenience
hygiene for the user.
The humans have ten ngers, if anything incident
happens to any one of the ngers, other ngers can
be used as a replacement for authentication7.
Finger veins can only be captured from an alive body.
Hence, if a person is dead, it is impossible to take his
identity 7.
Regardless of the advantages mentioned above, there are
defy that still should be dealt with in order to achieve the
higher performance wanted in the development of nger vein
biometric recognition. Firstly, the nger vein image acquisition
device has a good effect on the quality of the nger vein
images. The distance between the nger and camera is narrow
to one another, during the acquisition process. This closed
position can cause optical blurring on the captured image 8.
In addition, the very important feature into the system is the
lighting of the capturing device . Bad lighting may cause the
image to appear extremely dark or extremely bright 7.Besides
that, the position guide of the nger is also important. If the
nger is not guided, the recognition rate may be reduced as the
nger images could have an incorrect position or alignment.
Thus, matching may be wrong. Other than that, the thickness of
skin and bones varies for every individual. For that reason, may
happens light scattering because the humans skin layer is not
coherent 7.The loud quality images need to be ameliorated
, so to take control those issues, usually nger vein recogni-
tion methods implemented complicated image preprocessing
algorithms to the system 7.
Beyond these advantages and disadvantages, the nger
vein recognition techniques are used in many applications
7. Among them; login authentication, door security controls,
personal computer security, physical access management. This
paper is organized as follows: Section II describes nger
vein authentication steps, which include image acquisition
and databases. Section III elaborates existing approaches for
preprocessing, feature extraction and classication, and the
unsolved key problems of nger vein recognition are analyzed

TABLE I: COMPARISON BETWEEN BIOMETRIC SYSTEMS 6 .
Biometric Main Advantage Disadvantage Security Level Sensor Cost
Voice Natural and convenient Noise Normal Non contact Low
Face Remote capture Lighting conditions Normal Non contact Low
Fingerprint Widely applied Skin Good contact Low
Iris High precision Glasses Excellent Non contact High
Finger vein High security level Few Excellent Non contact Low
in section IV named discussions. Section V is conclusion of
the paper.
II. MAT E R I A L S A N D M E T H O D S
A. FINGER VEIN IDENTIFICATION AUTHENTICATION
STEPS
The verication of a person identity is very complex task
because we cannot identify the uniqueness of a person easily.
Biometric samples are not matched from raw data. Biometric
systems acquire raw data from which they extract key features.
As shown Fig. 1 a typical nger vein recognition system
mainly includes image acquisition, preprocessing, and feature
extraction and matching for classication 3. Fig. 1: A typical nger vein recognition system. B. Image Acquisition and databases
There are two ways that has been introduced for nger vein
image acquisition: light transmission method and light reec-
tion method 9.The main divergence between the methods is
the near-infrared light (NIR light) position. As illustrated in
Fig. 2 (a), in transmission principle imaging the light sources
is positioned in dorsal side of nger, and the image sensor
collects the images on the ventral side where the NIR light
will go through the nger. Contrariwise, in reection principal
imaging the light source is placed in palmer side of nger and
the images are collected through the reection of the NIR light,
as shown in Fig.2 (b). Transmission method comparing to light
reection method can capture a high contrast image, that why
most acquisition devices use a light transmission method 9,
10. Fig. 2: Finger vein principle imaging: (a) light transmission
method, (b) light reection method.
One of the greatest challenges to researchers is to get high
image quality, so there are varied public nger vein databases,
the ve typical databases are as follows:
The rst one it is was a part of homologous multimodal
databases is named SDUMLA-HMT database 11.Another
nger vein database which is also a homologous multimodal
databases known as HKPU-FV proposed by Ajay and Zhou
12 .The third database abbreviated as UTFV database was
realized by B.T. Ton, R.N.J. Veldhuis from University of
Twente 13. The tow nger vein databases were recently
published at Chonbuk Ntion University 14 and Tsinghua
University 15.the previous database are named respectively
MMCBNU-6000 database and THU-FVFDT database. The
TABLE 2 summarize mentioned databases all of them uses
light transmission based image acquisition device .but the

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

Fig. 3: Low image nger vein quality samples of SDUMLA-
HMT nger vein database.
number of subject/nger is limited , also the image size
,contrast ,backgrounds and quality are different. Some of the
samples are terribly skewed (misaligned) .Examples of low
image quality cases are shown in Fig. 3.
C. EXISTING APPROACHES FOR PREPROCESS-
ING,FEATURE EXTRACTION AND CLASSIFICAION
Generally, image captured by acquisition process itself
introduces ” noise “. This images need to be processed. More-
over, two different sensors of the same object will produce
different images 1. Preprocessing play an important role
to upgrade the quality of the captured image and founded
on the rules that dene how the blood vessel network will
be segmented .Finger vein preprocessing basically includes:
extraction of Region of Interest (ROI), segmentation and
combination of lters as reported by 11, 15 .The second
phase of the processing is to extract from each image most
useful or most relevant attribute. Characteristic extraction is
the key process in nger vein recognition, it reduce data and
transformed into set of features known as feature extraction.
Finger vein recognition system is divided into three cate-
gories according to the feature extraction methods. The rst
approach conventional method is also is known as a vein
pattern-based method . The second approach is that is named
dimensionality reduction-based methods by applying machine
learning technique . The nal approach is the local binary-
based method 7.
1) Vein-pattern based methods: There are typically six vein
pattern based feature extraction methods, maximum curvature
16, mean curvature 17, Gabor 12, 18, morphological
19, modied repeated line tracking 21 and line tracking
20.This group is the based method in nger vein extraction.
In this group, the vein patterns are rstly segmented,and then
the geometric feature or topological structure of vein pattern
is used for matching.
Maximum curvature, repeated line tracking, line tracking
all use cross-section of vein pattern looks like a valley, these
methods make based on this point to extract vein features from
images, but the particular methods of recognizing vein pixel
are dissimilar . However, mean curvature outlook the intensity
surface of nger vein image as a geometric object, and vein
pattern will be extracted from pixels with negative mean
curvature. Various from the above methods, which extract vein
pattern in the spatial domain, Gabor transforms extract vein
pattern image into the frequency domain. The vein patterns,
extracted by these group methods are binary, so the matched
pixel ratio is general used in matching 22. The matched pixel ratio or Accuracy means the ratio of the number of the
matching vein pixels to the total number of the vein pixels
in two vein patterns. The matching score is small because
the acquisition is non contact, nger displacement, nger
rotation and translation. Table ??summarizes the algorithms
applied by researchers for vein-pattern based methods and
comparison detail information about database there are two
family the public are general limited and proprietary are
developed in house . In Table ??also there are comparisons of
preprocessing, segmentation, feature extraction, classier and
EER (%) .
2) Dimensionality Reducction based Methods : Contrari-
wise learning methods usually transform image into low di-
mensional space to classify. Biometric traits captured a second
time is never exactly the same as the rst time so matching
has led to the usage of machine learning techniques, such as
Neural Networks and fuzzy logic. Machine learning possesses
key properties of being robust to noise and can efciently
solve complex pattern recognition problems. In nger vein
recognition, PCA 30, LDA PCA 23 and LBPV PCA 24
have been used. These methods need the training process to
learn a transformation matrix. When there are new enrolled
users, the transformation matrix need to learn again. So this
kind of methods may be not very practical. SVM classiers are
used in matching for these methods PCA 30, 23, 24. Table
IV lists out the algorithms applied in each existing machine
learning work.
3) Local Binary based Methods: Methods in this group are
based on local area, and the extracted features are in binary
formation. The Local Line Binary Pattern (LLBP)37, Local
Binary Pattern(LBP)33, Local Line Binary Pattern (LLBP)
38, Personalized Best Bit Map (PBBM)35 ,local directional
code (LDC)36 and personalized weight maps (PWM) 39
are all local binary methods .LLBP and LBP the local binary
code is obtained by compare the gray level of the current pixel
and its neighbors.PWM and PPBM further explore the stability
of binary codes ,and use the stable binary codes in matching
.Different from the four methods, LDC codes the local gradient
orientation information .For most of these methods, hamming
distance was used to compared between the enrolled and
input binary features’ of nger vein images. TABLEV show
more detailed comparison between some previously mentioned
Local binary based methods.
III. RE S U LT S A N D D I S C U S S I O N
Although, the improvement advancement of nger vein in
biometric recognition there still some important problem. The
rst one problem is about acquisition. In public databases,
there are some common issues about image quality. As
showing in section II, there are databases images with: low
contrast, images blurring, excessive brightness, excessive dark
and stains 36. Thus, there is space for the performance im-
provement nger vein acquisition device. The second problem
is the price of nger vein acquisition device, that still high
now, which is one factor that limits the application of nger
vein recognition 7.
By the combination of the two problems described above,
the device with low price and high performance will vastly
promote the development of nger vein recognition. The third

TABLE II: Comparison between nger vein databases
database dated images
numbers subject
numbers nger
numbers
per subject image
number pre nger image size
(pixels) format Typical Image
SDUMLA-HMT 11 2010 3816 106 index, ring,
middle, of
both hands 6 320 x 240 .bmp
HKPU-FV 12 2010 6264 156 index, ring,
middle, of
left hand 12/6* 513 x 256 .bmp
UTFV 13 2013 1440 60 index, ring,
middle, of
both hands 4 672 x 380 8 bit gray
scale .png MMCBNU-6000 14 2013 6000 100 index, ring,
middle, of
both hands 10 640 x 480 .bmp
THU-FVFDT 15 2014 440 220 2 1 200 x 100 .bmp
* For the second imaging session, there are 105 subjects changing, so each of ngers from them has 6 images, but others each have 12 images.
TABLE III: Vein-pattern based methods of nger vein recognition Reference Database Preprocessing Segmentation, feature extraction Classier Accuracy
/EER(%) XIANG.YANG 2009
20 Public ROI extraction Local threshold
Repeated line tracking matching –
Wonseok
Taejeong2011 19 Public Edge detection
ROI extraction
Smoothing lter mean curvature matching 0.25
Ajay Kumar2012 12 Proprietary ROI extraction
segmentation Gabor Filters
and Morphological Processing matching 1.22
Yang et al., 2013 22 Proprietary Anisotropic diffusion method
Non-scatter transmission maps
Gabor wavelet Directional ltering method Phase only correlation
strategy 0.0462
Perez Vega et al.,
2014 27 Public Edge detection ROI extraction
Smoothing lter Personalized Best Bit Map
(PBBM) Cross-correlation
matching 27.56
Lu et al., 2014 28 Public ROI extraction Histogram of Competitive
Gabor Responses (HCGR) Matching 0.671
Gupta and Gupta 2015
17 Public ROI extraction
Multiscale matched ltering
Line tracking Variation approach Sum of square
differences 4.47
M. Vlachos and E.
Dermatas. 2015 18 Proprietary ROI extraction
Brightness normalization
Minimization of the Mumford
Shah Model Local entropy thresholding
Morphological Dilation
Morphological ltering Template matching 24.65
M. Vlachos and E.
Dermatas. 2015 18 Proprietary ROI extraction
Brightness normalization
Minimizationof the Mumford
Shah Model Local entropy thresholding
Morphological Dilation
Morphological ltering Template matching 24.65
Fotios ,Vlachakis
2017 29 Public histogram equalization
brightness normalization
wiener ltering Local enhancement binarization
non-vein areas detection
whitening non-vein areas Template matching 0.5
Jinfeng ,Yihua 2017
16 Public Curve model
Spatial curve lter Weighted spatial curve lter
2D Gausian model matching 0.02

TABLE IV: Dimensionality reduction-based methods of nger vein recognition
Reference Database Preprocessing Segmentation, feature extraction Classier Accuracy
/EER(%) Wu and Liu, 2011
30 Proprietary ROI extraction
Image resize PCA ANFIS (neuro-fuzzy
system) 99
Wu and Liu, 2011
23 Proprietary ROI extraction
Image resize PCA
LDA SVM 98
Hosyar etal., 2011
26 Proprietary ROI extraction
Median ltering Histogram
equalization Morphological operation
Maximum curvature points MLP 93
Kuan-Quan et al.,
201224 Proprietary Gaussian matched lter Local Binary Pattern Variance
(LBPV) Global Matching SVM 79.00
Souad et al., 2014 26 Public Histogram equalization
Contrast amelioration
Median ltering
Gabor lter Global thresholding
Gabor lter SVM 98.75
Syafeeza etal.,2016
31 Proprietary ROI extraction
Image resize Binarized Image Convolutional Neural
Network 99.83
Hyung, lee 2017 32 Public ROI extraction Difference between input and
enrolled images convolutional neural
network (CNN) 96
TABLE V: Local binary-based method of nger vein recognition
Reference Database Preprocessing Segmentation, feature extraction Classier Accuracy
/EER(%) Rosdi et al., 2011 33 Proprietary Modied Gaussian high-pass lter Local Line Binary Pattern (LLBP) Hamming Distance
(HD) 1.78
Lee et al., 2011 34 Proprietary Gaussian high-pass lter Simple Binarization
ocal Binary Pattern(LBP)
LDP LDA Hamming Distance
(HD) 2.32
Xianjing, Gongping
2012 35 Proprietary ROI extraction
size normalization Local Directional Code LDC 100
Yang, Xiao 2013 36 Proprietary ROI extraction
size normalization
Gray normalization personalized weight maps (PWM) PWM-LLBP 99.67
Lu, Yoon 2013 37 Public contrast-limited adaptive
histogram equalization (CLAHE) Local Line Binary Pattern (LLBP) similarity between two
histograms 99.21
Perez Vega ,al., 2014
38 Proprietary Edge detection
ROI extraction
Smoothing lter Personalized Best Bit Map
(PBBM) Cross-correlation
matching 27.56
problem is nger displacement during acquisition. Since the
nger is a kind of non-rigid object and is arranged in a
non-contact way during image capturing 36, the nger vein
images are adversely affected by various deformations such
as: in-plane translations and rotations, out-of-plane rotations
and global or local expansion or contraction due to muscle
tension and joint states ?.Xianjing and Xiaoming ? pro-
posed methods based on matching operation, and extracted
pixel-based displacements to represent various deformations,
after which the uniformity of the displacement elds was used
to discriminate between genuine and imposter matching. Lee
et al 33 propose nger guidance to solve the displacement
problem.
Last on is the complexity of nger vein recognition algo-
rithm. So, this problem involved implantation systems. Kuk
and Lee 36 proposed development of human identication
system based on simple nger vein pattern matching method
for embedded environments. There are many remains to be
done on nger vein recognition to further improve its perfor-
mance, and promote its practical application.
IV. CO N C L U S I O N
In this paper, we review the recent development of nger
vein authentication, and give some typical works in this eld.
We have surveyed on existing databases. There are public
available databases and there are some works developed their own databases in house. The highlight of this paper is to
analyze a signicant number of papers to cover the existing
approach of nger vein recognition .In particular, we focus on
the technique employed in image acquisition, preprocessing,
feature extraction and classication. Besides, we have dis-
cussed some key problems of nger vein recognition. This
technique has a high potential to be the future research
direction in biometric recognition.
RE F E R E N C E S
1R. F. Amine Nait-Ali, Signal and image processing for biometrics .
2T. hafes, “Reconnaissance biométrique multimodale basée sur la fusion en score de deux modalités biométriques: l’empreinte digitale et la
signature.”
3C. Wilson, Vein pattern recognition: a privacy-enhancing biometric .
CRC press, 2010.
4S. Damavandinejadmonfared, A. K. Mobarakeh, S. A. Suandi, and B. A. Rosdi, “Evaluate and determine the most appropriate method to identify
nger vein,” Procedia Engineering , vol. 41, pp. 516–521, 2012.
5I. Claude, J.-L. Daire, and G. Sebag, “Fetal brain mri: segmentation and biometric analysis of the posterior fossa,” IEEE Transactions on
Biomedical Engineering , vol. 51, no. 4, pp. 617–626, 2004.
6L. Ballard, D. Lopresti, and F. Monrose, “Evaluating the security of handwriting biometrics,” in Tenth International Workshop on Frontiers
in Handwriting Recognition , Suvisoft, 2006.
7K. Syazana-Itqan, A. Syafeeza, N. Saad, N. A. Hamid, and W. H. B. M. Saad, “A review of nger-vein biometrics identication approaches,”
Indian Journal of Science and Technology , vol. 9, no. 32, 2016.

A review of nger vein biometrics authentication
System
Sara Daas
Laboratory L.E.R.I.C.A Electronics Department
University of Badji Mokhtar Annaba , Email:[email protected] Mohamed Boughazi
Laboratory L.E.R.I.C.A Electronics Department
University of Badji Mokhtar Annaba
Abstract —In nowadays, with the fast development in the eld
of electronic information technology, the identity verication and
security of data is a critical main problem. Due to this fact, the
biometric recognition has gaining popularity as it provides a high
security , reliable and robust approach for personal identica-
tion. As a new biometric technique, nger vein recognition has
attracted lots of attention from researchers. In this paper a survey
has been done on nger vein biometrics recognition taken from
existing research work on image preprocessing, feature extraction,
classication stage and the performance parameter of biometric
system the Equal Error Rate (EER).The main aim of this study
is to enlighten nger vein authentication research domain.
Keywords —Biometrics, nger vein recognition, preprocessing,
feature extraction, classication, EER .
I. IN T RO D U C T I O N
Biometrics is an emerging technique that enables us to
verify the identity of an individual by using at least one of
his or her personal characteristics 1.The biometric systems
are classied into three categories .biometric systems that are
based on behavioral patterns such as handwriting and keystroke
dynamics, that are based on physiological patterns such as
ngerprint, face and iris and the systems that are based on
hidden patterns such as the electrocardiogram (ECG) and MRI
images 2.Those systems somehow have several inconvenient;
the features with age may not be clear for identication and can
be easily stolen, lost or borrowed 3.A pattern of this type of
biometric trait is ngerprint and voice 1 ,also iris recognition
is considered as least user cause it is uneasiness to human eyes
due to the brightness of light during the biometric capture
process 4.MRI and ECG are expensive and complicated
techniques 5, 6 .The handwriting is a natural and familiar
way of conrming identity but not secure6 . TABLE.I show
more detailed comparison between some previously mentioned
biometric systems.
Recently in the literature, a biometric method based on
vascular patterns such as hand vein and peculiarly nger vein
has interested the researchers. A nger vein biometric system
came into existence after the invention of nger print. The
veins are within of the human nger. Finger vein features
offering varied other favorite advantages and these include:
The nger vein patterns are exceptional for every
person, even identical twins. So, it offers a great
distinction between each individual.
The patterns of ngers veins do not change with time,
they are permanent.
The nger vein patterns are invisible to humans eyes.
Consequently, they are not obscure and difcult to be
replicated of the fact that it is positioned underneath
the human skin.
The nger-vein patterns acquisition is considered to
be easy to use . The vein pattern images captured non
invasive. The nger vein device is contactless sensors,
so his concept ensuring hygiene and convenience
hygiene for the user.
The humans have ten ngers, if anything incident
happens to any one of the ngers, other ngers can
be used as a replacement for authentication7.
Finger veins can only be captured from an alive body.
Hence, if a person is dead, it is impossible to take his
identity 7.
Regardless of the advantages mentioned above, there are
defy that still should be dealt with in order to achieve the
higher performance wanted in the development of nger vein
biometric recognition. Firstly, the nger vein image acquisition
device has a good effect on the quality of the nger vein
images. The distance between the nger and camera is narrow
to one another, during the acquisition process. This closed
position can cause optical blurring on the captured image 8.
In addition, the very important feature into the system is the
lighting of the capturing device . Bad lighting may cause the
image to appear extremely dark or extremely bright 7.Besides
that, the position guide of the nger is also important. If the
nger is not guided, the recognition rate may be reduced as the
nger images could have an incorrect position or alignment.
Thus, matching may be wrong. Other than that, the thickness of
skin and bones varies for every individual. For that reason, may
happens light scattering because the humans skin layer is not
coherent 7.The loud quality images need to be ameliorated
, so to take control those issues, usually nger vein recogni-
tion methods implemented complicated image preprocessing
algorithms to the system 7.
Beyond these advantages and disadvantages, the nger
vein recognition techniques are used in many applications
7. Among them; login authentication, door security controls,
personal computer security, physical access management. This
paper is organized as follows: Section II describes nger
vein authentication steps, which include image acquisition
and databases. Section III elaborates existing approaches for
preprocessing, feature extraction and classication, and the
unsolved key problems of nger vein recognition are analyzed

TABLE I: COMPARISON BETWEEN BIOMETRIC SYSTEMS 6 .
Biometric Main Advantage Disadvantage Security Level Sensor Cost
Voice Natural and convenient Noise Normal Non contact Low
Face Remote capture Lighting conditions Normal Non contact Low
Fingerprint Widely applied Skin Good contact Low
Iris High precision Glasses Excellent Non contact High
Finger vein High security level Few Excellent Non contact Low
in section IV named discussions. Section V is conclusion of
the paper.
II. MAT E R I A L S A N D M E T H O D S
A. FINGER VEIN IDENTIFICATION AUTHENTICATION
STEPS
The verication of a person identity is very complex task
because we cannot identify the uniqueness of a person easily.
Biometric samples are not matched from raw data. Biometric
systems acquire raw data from which they extract key features.
As shown Fig. 1 a typical nger vein recognition system
mainly includes image acquisition, preprocessing, and feature
extraction and matching for classication 3. Fig. 1: A typical nger vein recognition system. B. Image Acquisition and databases
There are two ways that has been introduced for nger vein
image acquisition: light transmission method and light reec-
tion method 9.The main divergence between the methods is
the near-infrared light (NIR light) position. As illustrated in
Fig. 2 (a), in transmission principle imaging the light sources
is positioned in dorsal side of nger, and the image sensor
collects the images on the ventral side where the NIR light
will go through the nger. Contrariwise, in reection principal
imaging the light source is placed in palmer side of nger and
the images are collected through the reection of the NIR light,
as shown in Fig.2 (b). Transmission method comparing to light
reection method can capture a high contrast image, that why
most acquisition devices use a light transmission method 9,
10. Fig. 2: Finger vein principle imaging: (a) light transmission
method, (b) light reection method.
One of the greatest challenges to researchers is to get high
image quality, so there are varied public nger vein databases,
the ve typical databases are as follows:
The rst one it is was a part of homologous multimodal
databases is named SDUMLA-HMT database 11.Another
nger vein database which is also a homologous multimodal
databases known as HKPU-FV proposed by Ajay and Zhou
12 .The third database abbreviated as UTFV database was
realized by B.T. Ton, R.N.J. Veldhuis from University of
Twente 13. The tow nger vein databases were recently
published at Chonbuk Ntion University 14 and Tsinghua
University 15.the previous database are named respectively
MMCBNU-6000 database and THU-FVFDT database. The
TABLE 2 summarize mentioned databases all of them uses
light transmission based image acquisition device .but the

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

Fig. 3: Low image nger vein quality samples of SDUMLA-
HMT nger vein database.
number of subject/nger is limited , also the image size
,contrast ,backgrounds and quality are different. Some of the
samples are terribly skewed (misaligned) .Examples of low
image quality cases are shown in Fig. 3.
C. EXISTING APPROACHES FOR PREPROCESS-
ING,FEATURE EXTRACTION AND CLASSIFICAION
Generally, image captured by acquisition process itself
introduces ” noise “. This images need to be processed. More-
over, two different sensors of the same object will produce
different images 1. Preprocessing play an important role
to upgrade the quality of the captured image and founded
on the rules that dene how the blood vessel network will
be segmented .Finger vein preprocessing basically includes:
extraction of Region of Interest (ROI), segmentation and
combination of lters as reported by 11, 15 .The second
phase of the processing is to extract from each image most
useful or most relevant attribute. Characteristic extraction is
the key process in nger vein recognition, it reduce data and
transformed into set of features known as feature extraction.
Finger vein recognition system is divided into three cate-
gories according to the feature extraction methods. The rst
approach conventional method is also is known as a vein
pattern-based method . The second approach is that is named
dimensionality reduction-based methods by applying machine
learning technique . The nal approach is the local binary-
based method 7.
1) Vein-pattern based methods: There are typically six vein
pattern based feature extraction methods, maximum curvature
16, mean curvature 17, Gabor 12, 18, morphological
19, modied repeated line tracking 21 and line tracking
20.This group is the based method in nger vein extraction.
In this group, the vein patterns are rstly segmented,and then
the geometric feature or topological structure of vein pattern
is used for matching.
Maximum curvature, repeated line tracking, line tracking
all use cross-section of vein pattern looks like a valley, these
methods make based on this point to extract vein features from
images, but the particular methods of recognizing vein pixel
are dissimilar . However, mean curvature outlook the intensity
surface of nger vein image as a geometric object, and vein
pattern will be extracted from pixels with negative mean
curvature. Various from the above methods, which extract vein
pattern in the spatial domain, Gabor transforms extract vein
pattern image into the frequency domain. The vein patterns,
extracted by these group methods are binary, so the matched
pixel ratio is general used in matching 22. The matched pixel ratio or Accuracy means the ratio of the number of the
matching vein pixels to the total number of the vein pixels
in two vein patterns. The matching score is small because
the acquisition is non contact, nger displacement, nger
rotation and translation. Table ??summarizes the algorithms
applied by researchers for vein-pattern based methods and
comparison detail information about database there are two
family the public are general limited and proprietary are
developed in house . In Table ??also there are comparisons of
preprocessing, segmentation, feature extraction, classier and
EER (%) .
2) Dimensionality Reducction based Methods : Contrari-
wise learning methods usually transform image into low di-
mensional space to classify. Biometric traits captured a second
time is never exactly the same as the rst time so matching
has led to the usage of machine learning techniques, such as
Neural Networks and fuzzy logic. Machine learning possesses
key properties of being robust to noise and can efciently
solve complex pattern recognition problems. In nger vein
recognition, PCA 30, LDA PCA 23 and LBPV PCA 24
have been used. These methods need the training process to
learn a transformation matrix. When there are new enrolled
users, the transformation matrix need to learn again. So this
kind of methods may be not very practical. SVM classiers are
used in matching for these methods PCA 30, 23, 24. Table
IV lists out the algorithms applied in each existing machine
learning work.
3) Local Binary based Methods: Methods in this group are
based on local area, and the extracted features are in binary
formation. The Local Line Binary Pattern (LLBP)37, Local
Binary Pattern(LBP)33, Local Line Binary Pattern (LLBP)
38, Personalized Best Bit Map (PBBM)35 ,local directional
code (LDC)36 and personalized weight maps (PWM) 39
are all local binary methods .LLBP and LBP the local binary
code is obtained by compare the gray level of the current pixel
and its neighbors.PWM and PPBM further explore the stability
of binary codes ,and use the stable binary codes in matching
.Different from the four methods, LDC codes the local gradient
orientation information .For most of these methods, hamming
distance was used to compared between the enrolled and
input binary features’ of nger vein images. TABLEV show
more detailed comparison between some previously mentioned
Local binary based methods.
III. RE S U LT S A N D D I S C U S S I O N
Although, the improvement advancement of nger vein in
biometric recognition there still some important problem. The
rst one problem is about acquisition. In public databases,
there are some common issues about image quality. As
showing in section II, there are databases images with: low
contrast, images blurring, excessive brightness, excessive dark
and stains 36. Thus, there is space for the performance im-
provement nger vein acquisition device. The second problem
is the price of nger vein acquisition device, that still high
now, which is one factor that limits the application of nger
vein recognition 7.
By the combination of the two problems described above,
the device with low price and high performance will vastly
promote the development of nger vein recognition. The third

TABLE II: Comparison between nger vein databases
database dated images
numbers subject
numbers nger
numbers
per subject image
number pre nger image size
(pixels) format Typical Image
SDUMLA-HMT 11 2010 3816 106 index, ring,
middle, of
both hands 6 320 x 240 .bmp
HKPU-FV 12 2010 6264 156 index, ring,
middle, of
left hand 12/6* 513 x 256 .bmp
UTFV 13 2013 1440 60 index, ring,
middle, of
both hands 4 672 x 380 8 bit gray
scale .png MMCBNU-6000 14 2013 6000 100 index, ring,
middle, of
both hands 10 640 x 480 .bmp
THU-FVFDT 15 2014 440 220 2 1 200 x 100 .bmp
* For the second imaging session, there are 105 subjects changing, so each of ngers from them has 6 images, but others each have 12 images.
TABLE III: Vein-pattern based methods of nger vein recognition Reference Database Preprocessing Segmentation, feature extraction Classier Accuracy
/EER(%) XIANG.YANG 2009
20 Public ROI extraction Local threshold
Repeated line tracking matching –
Wonseok
Taejeong2011 19 Public Edge detection
ROI extraction
Smoothing lter mean curvature matching 0.25
Ajay Kumar2012 12 Proprietary ROI extraction
segmentation Gabor Filters
and Morphological Processing matching 1.22
Yang et al., 2013 22 Proprietary Anisotropic diffusion method
Non-scatter transmission maps
Gabor wavelet Directional ltering method Phase only correlation
strategy 0.0462
Perez Vega et al.,
2014 27 Public Edge detection ROI extraction
Smoothing lter Personalized Best Bit Map
(PBBM) Cross-correlation
matching 27.56
Lu et al., 2014 28 Public ROI extraction Histogram of Competitive
Gabor Responses (HCGR) Matching 0.671
Gupta and Gupta 2015
17 Public ROI extraction
Multiscale matched ltering
Line tracking Variation approach Sum of square
differences 4.47
M. Vlachos and E.
Dermatas. 2015 18 Proprietary ROI extraction
Brightness normalization
Minimization of the Mumford
Shah Model Local entropy thresholding
Morphological Dilation
Morphological ltering Template matching 24.65
M. Vlachos and E.
Dermatas. 2015 18 Proprietary ROI extraction
Brightness normalization
Minimizationof the Mumford
Shah Model Local entropy thresholding
Morphological Dilation
Morphological ltering Template matching 24.65
Fotios ,Vlachakis
2017 29 Public histogram equalization
brightness normalization
wiener ltering Local enhancement binarization
non-vein areas detection
whitening non-vein areas Template matching 0.5
Jinfeng ,Yihua 2017
16 Public Curve model
Spatial curve lter Weighted spatial curve lter
2D Gausian model matching 0.02

TABLE IV: Dimensionality reduction-based methods of nger vein recognition
Reference Database Preprocessing Segmentation, feature extraction Classier Accuracy
/EER(%) Wu and Liu, 2011
30 Proprietary ROI extraction
Image resize PCA ANFIS (neuro-fuzzy
system) 99
Wu and Liu, 2011
23 Proprietary ROI extraction
Image resize PCA
LDA SVM 98
Hosyar etal., 2011
26 Proprietary ROI extraction
Median ltering Histogram
equalization Morphological operation
Maximum curvature points MLP 93
Kuan-Quan et al.,
201224 Proprietary Gaussian matched lter Local Binary Pattern Variance
(LBPV) Global Matching SVM 79.00
Souad et al., 2014 26 Public Histogram equalization
Contrast amelioration
Median ltering
Gabor lter Global thresholding
Gabor lter SVM 98.75
Syafeeza etal.,2016
31 Proprietary ROI extraction
Image resize Binarized Image Convolutional Neural
Network 99.83
Hyung, lee 2017 32 Public ROI extraction Difference between input and
enrolled images convolutional neural
network (CNN) 96
TABLE V: Local binary-based method of nger vein recognition
Reference Database Preprocessing Segmentation, feature extraction Classier Accuracy
/EER(%) Rosdi et al., 2011 33 Proprietary Modied Gaussian high-pass lter Local Line Binary Pattern (LLBP) Hamming Distance
(HD) 1.78
Lee et al., 2011 34 Proprietary Gaussian high-pass lter Simple Binarization
ocal Binary Pattern(LBP)
LDP LDA Hamming Distance
(HD) 2.32
Xianjing, Gongping
2012 35 Proprietary ROI extraction
size normalization Local Directional Code LDC 100
Yang, Xiao 2013 36 Proprietary ROI extraction
size normalization
Gray normalization personalized weight maps (PWM) PWM-LLBP 99.67
Lu, Yoon 2013 37 Public contrast-limited adaptive
histogram equalization (CLAHE) Local Line Binary Pattern (LLBP) similarity between two
histograms 99.21
Perez Vega ,al., 2014
38 Proprietary Edge detection
ROI extraction
Smoothing lter Personalized Best Bit Map
(PBBM) Cross-correlation
matching 27.56
problem is nger displacement during acquisition. Since the
nger is a kind of non-rigid object and is arranged in a
non-contact way during image capturing 36, the nger vein
images are adversely affected by various deformations such
as: in-plane translations and rotations, out-of-plane rotations
and global or local expansion or contraction due to muscle
tension and joint states ?.Xianjing and Xiaoming ? pro-
posed methods based on matching operation, and extracted
pixel-based displacements to represent various deformations,
after which the uniformity of the displacement elds was used
to discriminate between genuine and imposter matching. Lee
et al 33 propose nger guidance to solve the displacement
problem.
Last on is the complexity of nger vein recognition algo-
rithm. So, this problem involved implantation systems. Kuk
and Lee 36 proposed development of human identication
system based on simple nger vein pattern matching method
for embedded environments. There are many remains to be
done on nger vein recognition to further improve its perfor-
mance, and promote its practical application.
IV. CO N C L U S I O N
In this paper, we review the recent development of nger
vein authentication, and give some typical works in this eld.
We have surveyed on existing databases. There are public
available databases and there are some works developed their own databases in house. The highlight of this paper is to
analyze a signicant number of papers to cover the existing
approach of nger vein recognition .In particular, we focus on
the technique employed in image acquisition, preprocessing,
feature extraction and classication. Besides, we have dis-
cussed some key problems of nger vein recognition. This
technique has a high potential to be the future research
direction in biometric recognition.
RE F E R E N C E S
1R. F. Amine Nait-Ali, Signal and image processing for biometrics .
2T. hafes, “Reconnaissance biométrique multimodale basée sur la fusion en score de deux modalités biométriques: l’empreinte digitale et la
signature.”
3C. Wilson, Vein pattern recognition: a privacy-enhancing biometric .
CRC press, 2010.
4S. Damavandinejadmonfared, A. K. Mobarakeh, S. A. Suandi, and B. A. Rosdi, “Evaluate and determine the most appropriate method to identify
nger vein,” Procedia Engineering , vol. 41, pp. 516–521, 2012.
5I. Claude, J.-L. Daire, and G. Sebag, “Fetal brain mri: segmentation and biometric analysis of the posterior fossa,” IEEE Transactions on
Biomedical Engineering , vol. 51, no. 4, pp. 617–626, 2004.
6L. Ballard, D. Lopresti, and F. Monrose, “Evaluating the security of handwriting biometrics,” in Tenth International Workshop on Frontiers
in Handwriting Recognition , Suvisoft, 2006.
7K. Syazana-Itqan, A. Syafeeza, N. Saad, N. A. Hamid, and W. H. B. M. Saad, “A review of nger-vein biometrics identication approaches,”
Indian Journal of Science and Technology , vol. 9, no. 32, 2016.

x

Hi!
I'm Elaine!

Would you like to get a custom essay? How about receiving a customized one?

Check it out