SIViP

DOI 10.1007/s11760-016-0936-z

ORIGINAL PAPER

Fingerprint liveness detection using gradient-based

texture features
Zhihua Xia1 · Rui Lv1 · Yafeng Zhu1 ·
Peng Ji1 · Huiyu Sun2 · Yun-Qing Shi3

Received: 1 December 2015 / Revised: 16 June 2016 / Accepted: 8 July 2016

© Springer-Verlag London 2016

Abstract Fingerprint-based recognition systems have been
increasingly deployed in various applications nowadays.
However, the recognition systems can be spoofed by using an
accurate imitation of a live fingerprint such as an artificially
made fingerprint. In this paper, we propose a novel software-
based fingerprint liveness detection method which achieves
good detection accuracy. We regard the fingerprint liveness
detection as a two-class classification problem and construct
co-occurrence array from image gradients to extract features.
In doing so, the quantization operation is firstly conducted
on the images. Then, the horizontal and vertical gradients
at each pixel are calculated, and the gradients of large
absolute values are truncated into a reduced range. Finally,
the second-order and the third-order co-occurrence arrays are
constructed from the truncated gradients, and the elements
of the co-occurrence arrays are directly used as features.
The second-order and the third-order co-occurrence array
features are separately utilized to train support vector
machine classifiers on two publicly available databases used
in Fingerprint Liveness Detection Competition 2009 and
2011. The experimental results have demonstrated that the
features extracted with the third-order co-occurrence array
achieve better detection accuracy than that with the second-
B Zhihua Xia
xia_zhihua@163.com
1 Jiangsu Engineering Center of Network Monitoring, Jiangsu
Collaborative Innovation Center on Atmospheric
Environment and Equipment Technology, College of
Computer and Software, Nanjing University of Information
Science and Technology, Nanjing 210044, China
2
Department of Computer Science, New York University,
New York, NY 10012, USA
3
Department of Electrical and Computer Engineering,
New Jersey Institute of Technology, New York, NJ 07102,
USA
order co-occurrence array and outperform the state-of-the-art
methods.
Keywords Biometrics · Fingerprint liveness detection ·
Image texture · Image gradient · Co-occurrence array
1 Introduction
Fingerprint has been used as biometric characteristics in
forensic science and practice for hundreds of years. Nowa-
days in the information age, fingerprint earns even more
applications because of its uniqueness and easiness to collect.
However, fingerprint-based systems are vulnerable to spoof
attacks, as the spoof fingerprint can be easily made using
cheap materials such as gelatin and silicon [1,2]. The
attackers could clandestinely retrieve a user’s fingerprint and
make a spoof one to achieve illegal access. In another way,
a user may make a spoof fingerprint of himself to deceive
attendance system.
In order to address the above problem, many fingerprint
liveness detection techniques have been proposed to judge
whether a fingerprint image was captured from an actual
user or not. The existing methods can be divided into two
classes: hardware-based methods applied at the acquisition
stage and software-based methods applied at the processing
stage [1,2].
The hardware-based methods try to add new hardware
components to obtain life signs of the finger such as
temperature, pulse, pulse oximetry, conductivity, and blood
pressure. Then, spoof fingers can be discriminated from real
ones by analyzing these signs. The hardware-based solutions
can prevent spoof attacks to some degree. However, this
type of techniques requires the support of additional devices
which increases the overall expenses of the fingerprint

123

identification system. On the other hand, the software-based
methods add liveness detection capability to the existing
systems by updating the software architecture. The software-
based methods distinguish the live fingers from the spoof
ones by analyzing the images obtained from the existing
fingerprint sensors without the need of new hardware. This
type of methods is less expensive and is more flexible to
future adaptions.
In this paper, a novel software-based fingerprint liveness
detection method is proposed. We regard the liveness detec-
tion as a two-class classification problem, i.e., classifying
a test fingerprint image into either a living or a spoof
one.
The remainder of the paper is organized as follows. In
Sect. 2, we give more related works related to fingerprint
liveness detection. Section 3 describes the feature extraction
process. In Sect. 4, we present a brief introduction to the two
databases and the support vector machine and then conduct
comparative experiments and furthermore, we discuss the
simulation results. Finally, we conclude our work and discuss
further work in Sect. 5.
2 Related works
The software-based methods show that live and spoof
fingerprints can be discriminated by analyzing the features
extracted from fingerprint images. Abhyankar and Schuckers
[3] developed a method based on multi-resolution texture
features and local ridge frequency features. Their texture fea-
tures include: (1) the first-order features, i.e., energy, entropy,
median, and variance of the histogram and (2) the second-
order features, i.e., cluster shade and cluster prominence of
the co-occurrence matrix. Coli et al. [4] claimed that the high-
frequency details of the spoof fingerprint images were greatly
reduced and extracted features from the power spectrum
for the classification. Nikam and Agarwal proposed several
liveness detection methods based on the texture analysis
of the fingerprint images. The authors extracted many
distinguishable features through various texture measure
methods such as the Gabor filters [5], the curvelet transform
[6], the Ridgelet transform [7], the wavelet transform [8],
and the gray-level co-occurrence matrices combined with
the wavelet transform [9]. Jiang and Liu [10] proposed a
spoof fingerprint detection methods based on difference co-
occurrence matrix. All of these features can successfully
address the fingerprint liveness detection problem to some
degree. Jin et al. [11] proposed a fingerprint liveness detection
method based on band-selective Fourier spectrum. The
authors revealed that the live fingerprint images showed
stronger Fourier spectrum in the ring patterns than the spoof
ones and classified live and spoof fingerprint images by
analyzing the band-selective Fourier spectral energies. Tan
123
SIViP
and Schuckers [12] revealed that artificial fingers have a
distinct noise distribution due to the material’s properties
and extracted features by analyzing the noise distribution
along the valley structure. Ghiani et al. [13] utilized binarized
statistic image features which is quite similar to local
binary pattern and local phase quantization descriptor.
Gragnaniello et al. [13] proposed a local-descriptor based
methods to handle fingerprint liveness detection problem.
The features are extracted from both the spatial and frequency
domains. Jia et al. [14] considered that the multi-scale local
binary pattern (LBP) could reflect the texture of fingerprint
images more adequately than original LBP and proposed a
fingerprint liveness detection method by using two kinds
of multi-scale LBP. Their method achieves good detection
accuracy.
3 Feature extraction
In this paper, the fingerprint liveness detection is considered
as a two-class classification problem. Feature extraction is a
crucial step for classification problems [15–19]. Here, X =
(X i, j ) ∈ {0, . . . , 255}n 1 ×n 2 represents an 8-bit grayscale
image of the size n 1 × n 2 . The symbol X i, j denotes the
grayscale value of the pixel located at (i, j), and n 1 and
n 2 denote the height and width of the image, respectively.
The process of the feature extraction includes the following
four steps. Firstly, the image is quantized with a quantization
factor. Secondly, the horizontal and vertical gradients are
calculated from the adjacent quantized pixels. Thirdly, the
gradients of large absolute values are truncated into a reduced
range. Finally, the co-occurrence arrays are constructed from
the gradients. The elements of the co-occurrence arrays are
directly used as the discriminative features.
3.1 Quantization
In the process of feature extraction, the image pixel values
are firstly quantized as:
X i, j ← X i, j /Q (1)
where Q ≥ 1 is a quantization factor. The quantization
operation will cause the loss of image information, but will
not affect the overall texture of a fingerprint image. The
quantization operation can largely reduce the dynamic range
of X i, j , and thus helping to reduce the dimensionality of the
feature vector. Moreover, the experimental results presented
in Sect. 4 have revealed that the feature vector extracted from
the quantized images achieve better detection accuracy than
that from the unquantized ones when a suitable quantization
factor is used.
SIViP

Q=1 Q=2 Q=4

0.35 0.4 0.5

0.3
0.4
0.3
0.25
Percentage

Percentage

Percentage
0.2 0.3
0.2
0.15 0.2
0.1
0.1
0.1
0.05

0 0 0
−200 −100 0 100 200 −100 −50 0 50 100 −100 −50 0 50 100
Value of Gradient Value of Gradient Value of Gradient

Q=8 Q=16 Q=32

0.7 0.7 0.8

0.6 0.6
0.6
0.5 0.5
Percentage

Percentage

Percentage
0.4 0.4
0.4
0.3 0.3

0.2 0.2
0.2
0.1 0.1

0 0 0
−50 0 50 −20 0 20 −15 −10 −5 0 5 10 15
Value of Gradient Value of Gradient Value of Gradient

Fig. 1 The histograms of horizontal gradients calculated from quantized fingerprint images, Q = 1, 2, 4, 8, 16, 32

3.2 Image gradient and truncation operation
Image gradient measures a directional change in the intensity
of an image, which can be used for effective texture feature
extraction [20,21]. Here, the horizontal and vertical gradient
arrays G H and G V are defined by calculating the differences
between the adjacent pixels as:
⎧
⎪ G (i, j) = X i, j − X i, j+1 ,
⎪
⎪ H
⎪
⎨ if G(i, j) < −T , truncT (G(i, j)) = −T ; if G(i, j) ∈
for i ∈ {0, . . . , n 1 − 1}, j ∈ {0, . . . , n 2 − 2};
⎪ G V (i, j) = X i, j − X i+1, j ,
⎪
⎪
⎪
⎩ for i ∈ {0, . . . , n 1 − 2}, j ∈ {0, . . . , n 2 − 1}.
(2)
In our method, the elements of the co-occurrence array
are directly used as features which are arranged to com-
pose the feature vector. The dimensionality of the feature
vector depends on the dynamic range of the gradients. The
quantization operation will decrease the dynamic range of
gray values and accordingly decrease the dynamic range of
gradients. As shown in Fig. 1, the gradients calculated from
the quantized images have a narrower dynamic range than
that from the original images. A large quantization factor will
reduce the dynamic range of the gradients and thus help to
reduce the dimensionality of the feature vector. Specifically,
the histogram of the gradients can be approximated by
Laplacian distribution. Thus, we can truncate gradients to a
small range [−T, T ] without losing much useful information.
We can cover a major part of gradients with a smaller T after
quantizing the pixels with a larger quantization parameter Q.
The truncation operation is defined as:
G(i, j) ← truncT (G(i, j)) (3)
where G(i, j) before truncating denotes the gradient calcu-
lated according to Formula (2) along horizontal or vertical
direction. Here, if G(i, j) > T , truncT (G(i, j)) = T ;
[−T, T ], truncT (G(i, j)) = G(i, j).
3.3 Co-occurrence array
Co-occurrence matrix (array) is a widely used tool to identify
the texture of images. Often, gray-level co-occurrence
matrix (GLCM) is defined based on the gray values of
an image and is typically large in dimensionality [22].
For an 8-bit grayscale image, the general GLCM will
generate 256 × 256 = 65536 elements which can not be
directly used as features due to the too high dimensionality.
In order to extract a compact feature vector, various metrics
of the matrix, such as energy, entropy, contrast, ,and
correlation, are often further applied, such as shown in
[3]. In the proposed method, the co-occurrence arrays are
constructed from the gradients between adjacent pixels.
And the dimensionality of the co-occurrence array can be
efficiently reduced by quantization and truncation operations.
Thus, the elements of the co-occurrence array can be directly

123
 SIViP

used as features. In this paper, we calculate second-order Table 1 Dimensionality of the SCAG feature vector and average
and third-order co-occurrence arrays from the gradient array percentages of the covered gradients when different parameter pairs
(Q, T ) are chosen
G H and G V . Here, the second-order co-occurrence array is
parameterized from the pairs of adjacent gradients, and the (Q, T ) Dimensionality of the Average percentage of
third-order co-occurrence array is parameterized from the SCAG feature vector covered gradients (%)
triplets of adjacent gradients. Specifically, the second-order (1, 10) 882 70.73
co-occurrence arrays on gradients (SCAG) along horizontal (2, 10) 882 81.23
and vertical directions are defined as: (4, 10) 882 91.43
⎧ (8, 5) 242 90.67
n1 −1 n
2 −3
⎪
⎪
⎪
⎪ SCAGH (s, t) = ϕ(G H (i, j), s) (16, 4) 162 96.05
⎪
⎪
⎪
⎪ (32, 4) 162 99.74
⎪
⎪
i=0 j=0
⎪
⎨ × ϕ(G H (i, j + 1), t),
(4)
⎪
⎪ n1 −3 n
2 −1
⎪
⎪
⎪
⎪ SCAGV (s, t) = ϕ(G V (i, j), s) of elements in TCAG H and TCAGV is 2 × (2T + 1)3 , which
⎪
⎪
⎪
⎪ i=0 j=0 will increase much faster with a larger truncation factor T .
⎪
⎩
× ϕ(G V (i + 1, j), t), Thus, we further reduce the number of elements in the third-
order co-occurrence arrays by averaging the elements of the

1, if x = y arrays in the two directions. Specifically, for the third-order
where s, t ∈ {−T, . . . , T }, and ϕ(x, y) = . co-occurrence array, the final features are calculated as:
0, if x = y
And likewise, the third-order co-occurrence arrays on
gradients (TCAG) along horizontal and vertical directions TCAGH (s, t, r ) + TCAGV (s, t, r )
are defined as: TCAGs,t,r = (8)
2
⎧ n1 −1 n
2 −4
⎪
⎪
⎪
⎪ TCAG (s, t, r ) = ϕ(G H (i, j), s)
⎪
⎪
H We denote the feature vector constructed with the third-
⎪
⎪
⎪
⎪
i=0 j=0
order co-occurrence array as TCAG whose dimensionality is
⎪
⎪
⎪
⎪ × ϕ(G H (i, j + 1), t) reduced to (2T + 1)3 .
⎪
⎪
⎨ × ϕ(G H (i, j + 2), r ),
(5)
⎪
⎪ n1 −4 n
2 −1
⎪
⎪
⎪
⎪ TCAGV (s, t, r ) = ϕ(G V (i, j), s)
⎪
⎪ 4 Experiments
⎪
⎪ i=0 j=0
⎪
⎪
⎪
⎪ × ϕ(G V (i + 1, j), t)
⎪
⎪ In our experiment, the average classification error (ACE) is
⎩
× ϕ(G V (i + 2, j), r ), defined as the evaluation criterion:

where s, t, r ∈ {−T, . . . , T }. In order to eliminate the effect

caused by the size of the image, the elements of the co- ACE = (FAR + FRR)/2 (9)
occurrence arrays are normalized as:
⎧
⎪ SCAGH (s, t) where false accept rate (FAR) is the proportion of spoof
⎪
⎪ SCAGH (s, t) ← T T ,
⎨
s=−T t=−T SCAGH (s, t) fingerprints being incorrectly accepted, and false reject rate
(6)
⎪
⎪ SCAGV (s, t) (FRR) is the proportion of real fingerprints being incorrectly
⎪
⎩ SCAGV (s, t) ← T T .
s=−T t=−T SCAGV (s, t) rejected.
⎧
⎪ TCAGH (s, t, r ) The performances achieved by using the SCAG and TCAG
⎪
⎪ TCAGH (s, t, r ) ← T T T ,
⎨ are separately tested on two databases named LivDet09DB
s=−T t=−T r =−T TCAGH (s, t, r )
(7) [23] and LivDet11DB [24]. In this section, we firstly present
⎪
⎪ TCAGV (s, t, r )
⎪
⎩ TCAGV (s, t, r ) ← T T T .
a brief introduction to the two databases and the support
s=−T t=−T r =−T TCAGV (s, t, r )
vector machine (SVM). Secondly, we conduct comparative
For the second-order co-occurrence array, all the elements experiments on LivDet11DB to choose a proper quantization
of SCAG H and SCAGV are used as features. We denote factor Q and a proper truncation factor T . Finally, the
the feature vector constructed with the second-order co- performance of the proposed method with the chosen
occurrence array as SCAG whose dimensionality is 2×(2T + parameters is compared with that achieved by the state-of-
1)2 . For the third-order co-occurrence array, the total number the-art works.

123
SIViP
4.1 Databases
The proposed detection methods are tested on two databases,
named LivDet09DB and LivDet11DB. The LivDet09DB
is the database used in Fingerprint Liveness Detection
Competition 2009 [23]. It consists of fingerprint images
taken from three different sensors: Biometrika, Identix,
and Crossmatch. The spoof fingers are generated using
three different materials: silicone, gelatin, and playdoh. The
LivDet11DB is the database used in Fingerprint Liveness
Detection Competition 2011 [24]. It consists of images
from four different sensors: Biometrika, Digital Persona,
Italdata, and Sagem. The spoof fingers are generated using six
different materials: gelatin, latex, playdoh, silicone, ecoflex,
and wood glue. All of the fingerprint images are transformed
into gray images before used. Each of the datasets has been
divided into two non-overlapping parts: training and testing
sets, which are used respectively in the training and testing
processes of the classification. For the two databases, images
captured from each sensor are tested separately.
4.2 Support vector machine (SVM)
SVM is utilized to train the classifier with feature vector
in this paper. Treating the data as two sets of points in an
n-dimensional space, SVM builds an separating hyperplane
by Lagrangian multipliers to differentiate the negative data
points from the positive ones [25–28]. Intuitively, a good
separation is achieved when a separating hyperplane has the
largest distance to the boundary points of both classes.
LIBSVM [29] is a free software for support vector clas-
sification. LIBSVM implements four basic kernels, among
which the radial basis function (RBF) kernel, e−γ xi −x j  , is
2
widely suggested as the best choice by the users. LIBSVM
also provides a tool named “Cross-validation and Grid-
search” to search the appropriate penalty parameter C and
kernel parameter γ for an RBF kernel. With this tool, one can
firstly set a relatively large range to find the parameters. Next,
after running for several rounds, one can estimate where
the better parameters could be found and shrink the search
range to refine the search. In this paper, the LIBSVM with
Table 2 The average
Parameter Average classification error (ACE) (%)
classification error (ACE) of the
pair (Q, T )
SCAG feature vector calculated LivDet11DB#1
with different parameter pairs
(Q, T ) (1, 10) 15.70
(2, 10) 12.25
(4, 10) 8.95
(8, 5) 8.00
(16, 4) 8.45
(32, 4) 10.45
RBF kernel is used to train classifiers in the experiments,
and the tool “Cross-validation and Grid-search” is utilized
to search the penalty parameter C and kernel parameter γ .
In our experiment, most of parameters are found in [25 , 230 ]
for C, and in [2−5 , 210 ] for γ .
4.3 Quantization factor Q and truncation factor T
In order to obtain a compact feature vector and good
detection accuracy, the selection of the quantization factor
Q and truncation factor T requires proper attention. In this
subsection, we report the comparison experiment conducted
on LivDet11DB to figure out the relatively better choices.
First, we set Q = 1, 2, 4, 8, 16, 32 to quantize the images.
Under each of the quantization step, we compute the average
percentages of gradients that fall into the range [−T, T ],
which then help us to choose the truncation factor T .
In our method, the dimensionality of the SCAG feature
vector is equal to 2×(2T +1)2 . Generally, we want to take as
many gradients into consideration as possible in constructing
feature vector. However, a larger T will lead to a feature
vector of larger dimensionality. By compromising on the
percentages of the covered gradients and the dimensionality
of the SCAG feature vector, we choose T = 10 when Q is
set to 1, 2 and 4, T = 5 when Q is set to 8, and T = 4 when
Q is set to 16 and 32. For each parameter pair (Q, T ), the
dimensionality of the SCAG feature vector and the average
percentages of the covered gradients are listed in Table 1.
Note that it is not guaranteed that the parameters used here
are the best choices.
The dimensionality of the TCAG feature vector is equal to
(2T + 1)3 and will increase much faster with a larger T than
that of the SCAG feature vector. Therefore, we choose T = 3
for all Q = 1, 2, 4, 8, 16, 32. Thus, the dimensionalities of
all the TCAG feature vectors are equal to 343.
The different parameter pair (Q, T ) produces different
feature vectors. Each of these feature vectors is used to train
SVM classifier separately. The detection accuracies of the
classifiers are listed in Tables 2 and 3.
As shown in Table 2, the SCAG feature vectors extracted
with the parameter pairs (8, 5) and (16, 4) achieve the best
LivDet11DB#2 LivDet11DB#3 LivDet11DB#4 Average
15.40 16.50 5.50 15.78
17.50 22.35 5.45 14.39
13.65 20.95 4.91 12.12
14.15 16.90 4.86 10.98
15.35 14.85 5.26 10.98
15.90 18.15 7.56 13.02
123
 SIViP

Table 3 The average

Parameter Average classification error (ACE) (%)
classification error (ACE) of the
pair (Q, T )
TCAG feature vector calculated LivDet11DB#1 LivDet11DB#2 LivDet11DB#3 LivDet11DB#4 Average
with different parameter pairs
(Q, T ) (1, 3) 12.10 5.80 29.25 3.39 12.63
(2, 3) 8.40 5.00 26.40 4.13 10.98
(4, 3) 7.60 3.10 22.50 4.03 9.31
(8, 3) 6.45 3.80 13.45 4.13 6.96
(16, 3) 6.70 4.75 11.75 3.34 6.63
(32, 3) 9.10 6.50 12.75 4.91 8.32

average detection result. When the quantization factor Q
is set to 4, we need to set the truncation factor T to 10
so as to cover 91.43 % of gradients, which generates 882
features. When the quantization factor Q is set to 8, we need
to set the truncation factor T to 5 so as to cover 90.67 % of
gradients, which generates only 242 features. However, the
feature vector extracted with the parameter pair (8, 5) achieve
better detection accuracy than that with (4, 10). This tells us
that the quantization operation with a suitable factor will help
to extract better features and reduce the dimensionality of the
feature vector.
As shown in Table 3, the TCAG feature vectors extracted
with the parameter pair (16, 3) obtain the best average
detection result. In addition, the TCAG feature vectors
achieve overall better detection results than that of the
SCAG ones. It can be attributed to that the third-order co-
occurrence array can describe the gradient texture better than
the second-order one for the fingerprint liveness detection
purpose.
4.4 Comparison with previous methods
The fingerprint liveness detection problem has been rese-
arched for more than ten years, and many excellent methods
have been proposed to identify the vitality signs of fingerprint
with high accuracy. According to the results shown in Tables
2 and 3, we choose the TCAG feature vector with the
parameter pair (16, 3) to compare with the previous methods.
Ghiani et al. [30] investigated the performance of sev-
eral classical fingerprint liveness detection methods on
LivDet11DB in 2012. In 2014, Jia et al. [14] proposed an
excellent fingerprint liveness detection method based on
multi-scale local binary pattern and tested their method on
LivDet11DB. Thus, we cite the results reported in [30] and
[14] and the best results achieved in LivDet 2011 [24] for
performance comparison on LivDet11DB.
Galbally et al. [31] proposed an excellent fingerprint
liveness detection method based on image quality and
conducted the comparison experiments on LivDet09DB.
Here, we cite the results reported in [31] and the best results
achieved in LivDet 2009 [23] for performance comparison
on LivDet09DB. As shown in Tables 4 and 5, the proposed

Table 4 Performance comparison in terms of average classification error (ACE) on LivDet11DB

Methods Average classification error ACE (%)
LivDet11DB#1 LivDet11DB#2 LivDet11DB#3 LivDet11DB#4 Average

TCAG with (16,3) 6.7 4.75 11.75 3.34 6.635

MSLBP1 [14] 7.3 2.5 14.8 5.3 7.475
MSLBP2 [14] 10.6 6.7 12.6 5.6 8.875
MLBP [32] reported in [14] 10.8 7.1 16.6 6.4 10.225
Original LBP [33] reported in [14] 13.0 10.8 24.1 11.5 14.85
Tan’s method [34] reported in [14] 43.8 18.2 29.6 24.7 29.075
Valleys wavelet [12] reported in [30] 29.0 13.0 23.6 28.0 23.4
Curvelet GLCM [35] reported in [30] 22.9 18.3 30.7 28.0 24.975
Power spectrum [4] reported in [30] 30.6 27.1 42.8 31.5 33
Wavelet energy [8] reported in [30] 50.2 14.0 46.8 22.0 33.25
Curvelet energy [35] reported in [30] 45.2 21.9 47.9 28.5 35.875
Ridges wavelet [36] reported in [30] 38.8 27.5 56.9 20.5 35.925
Best result in LivDet 2011 [24] 20.0 36.1 21.8 13.8 22.925

123
SIViP

Table 5 Performance comparison in terms of average classification error (ACE) on LivDet09DB

Methods Average classification error ACE (%)
LivDet09DB#1 LivDet09DB#2 LivDet09DB#3 Average

TCAG with (16,3) 11.3 5.4 2.0 6.2

IQA-based [31] 12.8 10.7 1.2 8.2
Best result in LivDet 2009 [23] 18.2 9.4 2.8 10.1
Marasco et al. [37] 12.6 15.2 9.7 12.5
Moon et al. [38] reported in [37] 23.0 23.5 38.2 28.2
Nikam et al. [6] reported in [37] 28.3 18.7 30.3 25.8
Abhyankar et al. [3] reported in [37] 31.7 31.5 47.2 36.8

method achieves better results than the previous methods in ing (KJR1402), Fund of MOE Internet Innovation Platform (KJRP1403),
general. CICAEET, and PAPD fund.

5 Conclusion References

1. Sousedik, C., Busch, C.: Presentation attack detection methods for

In this paper, we regard the fingerprint liveness detection
as a two-class classification problem and present a novel
software-based fingerprint liveness detection method which
achieves good detection accuracy. For the first time, the
elements of co-occurrence array constructed from image
gradients have been used for the fingerprint liveness detection
purpose. In doing so, firstly, quantization operation is applied
to reduce the dynamic range of pixel value. Secondly, image
gradients are calculated from adjacent quantized pixels
along horizontal and vertical directions. Thirdly, the second-
order and third-order co-occurrence arrays are respectively
constructed from the truncated gradients. The elements
of the co-occurrence arrays are directly used as features.
The second-order and the third-order co-occurrence array
features are separately used to train SVM classifier on two
popularly utilized databases. The experimental results have
demonstrated that the proposed methods outperform many
state-of-the-art methods in general.
The proposed method may be further improved. Firstly,
image gradients are measured by the differences between
adjacent pixels in this paper, which is a simple method and
other sophisticated methods can be used to measure the image
gradient better. Secondly, using a combination of various
features that are calculated with different parameters is likely
to achieve a better performance. Thirdly, our experimental
results have shown that the third-order co-occurrence array
features achieve better detection accuracy than second-order
ones. Better results might be obtained by using higher order
co-occurrence arrays. However, one may face a larger feature
dimensionality. These will be our future research goals.
Acknowledgments This work is supported by the NSFC (61173141,
U1536206, 61232016, U1405254, 61373133, 61502242, 61572258),
BK20150925, Fund of Jiangsu Engineering Center of Network Monitor-
fingerprint recognition systems: a survey. IET Biom. 3(4), 219–233
(2014)
2. Al-Ajlan, A.: Survey on fingerprint liveness detection. In:
International Workshop on Biometrics and Forensics, pp. 1–5,
IEEE (2013)
3. Abhyankar, A., Schuckers, S.: Fingerprint liveness detection using
local ridge frequencies and multiresolution texture analysis tech-
niques. In: IEEE International Conference on Image Processing,
pp. 321–324, IEEE (2006)
4. Coli, P., Marcialis, G.L., Roli, F.: Power spectrum-based fingerprint
vitality detection. In: IEEE Workshop on Automatic Identification
Advanced Technologies, pp. 169–173, IEEE (2007)
5. Nikam, S.B., Agarwal, S.: Gabor filter-based fingerprint anti-
spoofing. In: Advanced concepts for intelligent vision systems, pp.
1103–1114, Springer (2008)
6. Nikam, S.B., Agarwal, S.: Curvelet-based fingerprint anti-
spoofing. Signal Image Video Process. 4(1), 75–87 (2010)
7. Nikam, S.B., Agarwal, S.: Ridgelet-based fake fingerprint
detection. Neurocomputing 72(10), 2491–2506 (2009)
8. Nikam S.B., Agarwal, S.: Texture and wavelet-based spoof
fingerprint detection for fingerprint biometric systems. In: First
International Conference on Emerging Trends in Engineering and
Technology, pp. 675–680, IEEE (2008)
9. Nikam, S.B., Agarwal, S.: Wavelet energy signature and
GLCM features-based fingerprint anti-spoofing. In: International
Conference on Wavelet Analysis and Pattern Recognition, vol. 2,
pp. 717–723, IEEE (2008)
10. Jiang, Y., Liu, X.: Spoof fingerprint detection based on co-
occurrence matrix. Int. J. Signal Process. Image Process. Pattern
Recognit. 8(8), 373–384 (2015)
11. Jin, C., Kim, H., Elliott, S.: Liveness detection of fingerprint based
on band-selective fourier spectrum. In: Nam, K.-H., Rhee, G. (eds.)
Information Security and Cryptology-ICISC 2007, pp. 168–179.
Springer, Berlin (2007)
12. Tan, B., Schuckers, S.: New approach for liveness detection in
fingerprint scanners based on valley noise analysis. J. Electron.
Imaging 17(1), 011009–011009 (2008)
13. Ghiani, L., Hadid, A., Marcialis, G.L., Roli, F.: Fingerprint liveness
detection using binarized statistical image features. In: IEEE Sixth
International Conference on Biometrics: Theory, Applications and
Systems, pp. 1–6, IEEE (2013)

123

14. Jia, X., Yang, X., Cao, K., Zang, Y., Zhang, N., Dai, R., Zhu,
X., Tian, J.: Multi-scale local binary pattern with filters for spoof
fingerprint detection. Inf. Sci. 268, 91–102 (2014)
15. Chen, B., Shu, H., Coatrieux, G., Chen, G., Sun, X., Coatrieux,
J.L.: Color image analysis by quaternion-type moments. J. Math.
Imaging Vis. 51(1), 124–144 (2014)
16. Li, J., Li, X., Yang, B., Sun, X.: Segmentation-based image copy-
move forgery detection scheme. IEEE Trans. Inf. Forensics Secur.
10(3), 507–518 (2015)
17. Pan, Z., Zhang, Y., Kwong, S.: Efficient motion and disparity
estimation optimization for low complexity multiview video
coding. IEEE Trans. Broadcast. 61(2), 1–1 (2015)
18. Zheng, Y., Byeungwoo, J., Xu, D., Wu, Q.M.J., Zhang, H.: Image
segmentation by generalized hierarchical fuzzy c-means algorithm.
J. Intell. Fuzzy Syst. 28(2), 4024–4028 (2015)
19. Wen, X., Shao, L., Xue, Y., Fang, W.: A rapid learning algorithm
for vehicle classification. Inf. Sci. 295, 395–406 (2015)
20. Wei, S.-D., Lai, S.-H.: Robust and efficient image alignment based
on relative gradient matching. IEEE Trans. Image Process. 15(10),
2936–2943 (2006)
21. Xia, Z., Wang, X., Sun, X., Liu, Q., Xiong, N.: Steganalysis of lsb
matching using differences between nonadjacent pixels. Multimed.
Tools Appl. 75(4), 1947–1962 (2016)
22. Xia, Z., Wang, X., Sun, X., Wang, B.: Steganalysis of least
significant bit matching using multi-order differences. Secur.
Commun. Netw. 7(8), 1283–1291 (2014)
23. Marcialis, G.L., Lewicke, A., Tan, B., Coli, P., Grimberg, D.,
Congiu, A., Tidu, A., Roli, F., Schuckers, S.: First international
fingerprint liveness detection competition–livdet 2009. In: Foggia,
P., Sansone, C., Vento, M. (eds.) Image Analysis and Processing-
ICIAP 2009, pp. 12–23. Springer, Berlin (2009)
24. D. Yambay, L., Ghiani, P., Denti, G.L., Marcialis, F., Roli, Schuck-
ers, S.: Livdet 2011-fingerprint liveness detection competition
2011. In: IAPR International Conference on Biometrics, pp. 208–
215, IEEE (2012)
25. Gu, B., Sheng, V.S., Tay, K.Y., Romano, W., Li, S.: Incremental
support vector learning for ordinal regression. IEEE Trans. Neural
Netw. Learn. Syst. 26(7), 1403–1416 (2014)
26. Gu, B., Sheng, V.S., Wang, Z., Ho, D., Osman, S., Li, S.:
Incremental learning for v-support vector regression. Neural Netw
67(C), 140–150(2015)
123
SIViP
27. Gu, B., Sheng, V.S.: A robust regularization path algorithm for
v-support vector classification. IEEE Trans. Neural Netw. Learn.
Syst. PP(99), 1–8 (2016)
28. Gu, B., Sun, X., Sheng, V.S.: Structural minimax probability
machine. IEEE Trans. Neural Netw. Learn. Syst. PP(99), 1–11
(2016)
29. Chang, C.-C., Lin, C.-J.: Libsvm: a library for support vector
machines. ACM Trans. Intell. Syst. Technol. 2(3), 27 (2011)
30. Ghiani, L., Denti, P., Marcialis, G.L.: Marcialis, experimental
results on fingerprint liveness detection. In: Perales, F.J., Fisher,
R.B., Moeslund, T.B. (eds.) Articulated Motion and Deformable
Objects, pp. 210–218. Springer, Berlin (2012)
31. Galbally, J., Marcel, S., Fierrez, J.: Image quality assessment for
fake biometric detection: application to iris, fingerprint and face
recognition. IEEE Trans. Image Process. 23, 710–724 (2014)
32. Ojala, T., Pietikainen, M., Maenpaa, T.: Multiresolution gray-
scale and rotation invariant texture classification with local binary
patterns. IEEE Trans. Pattern Anal. Mach. Intell. 24(7), 971–987
(2002)
33. Ojala, T., Pietikäinen, M., Harwood, D.: A comparative study of
texture measures with classification based on featured distributions.
Pattern Recognit. 29(1), 51–59 (1996)
34. Tan, B., Schuckers, S.: Spoofing protection for fingerprint scanner
by fusing ridge signal and valley noise. Pattern Recognit. 43(8),
2845–2857 (2010)
35. Nikam, S., Agarwal, S.: Fingerprint liveness detection using
curvelet energy and co-occurrence signatures. In: Fifth Inter-
national Conference on Computer Graphics, Imaging and
Visualisation, pp. 217–222, IEEE (2008)
36. Tan, B., Schuckers, S.: Liveness detection for fingerprint scanners
based on the statistics of wavelet signal processing. In: Conference
on Computer Vision and Pattern Recognition Workshop, pp. 26–26,
IEEE (2006)
37. Marasco, E., Sansone, C.: Combining perspiration-and
morphology-based static features for fingerprint liveness detection.
Pattern Recognit. Lett. 33(9), 1148–1156 (2012)
38. Moon, Y.S., Chen, J., Chan, K., So, K., Woo, K.: Wavelet based
fingerprint liveness detection. Electron. Lett. 41(20), 1112–1113
(2005)