ArticlePDF Available

Abstract and Figures

This paper introduces a novel method for the generation of high-resolution synthetic hand-print images. Specific traits, such as fingerprint, palmprint, and hand-shape, are synthesized to obtain a whole hand-print. Each trait is generated by a methodology that mimics the nature of the corresponding biometric data and their main degrees of freedom. The biometric traits are then integrated into a single high-resolution realistic image. A quantitative validation of the obtained patterns is carried out in the context of minutiae matching by comparing genuine and impostor distributions between synthetic and real hand-prints. The proposed approach also proved to be useful for algorithm training/optimization.
Content may be subject to copyright.
Abstract This work introduces a novel method for the
generation of high-resolution synthetic hand-print images.
Specific traits such as fingerprint, palmprint and hand-shape are
synthesized to obtain a whole hand-print. Each trait is generated
by a methodology that mimics the nature of the corresponding
biometric data and their main degrees of freedom. The biometric
traits are then integrated into a single high-resolution realistic
image. A quantitative validation of the obtained patterns is carried
out in the context of minutiae matching by comparing genuine and
impostor distributions between synthetic and real hand-prints.
The proposed approach also proved to be useful for algorithm
Index Terms Biometrics, hand-print, hand shape, palmprint,
iometrics plays an increasingly important role in
authentication and identification systems. Biometric
recognition enables identification of individuals based on
their physical or behavioral characteristics [1]. Many modalities
have been studied, such as fingerprint, face, iris, voice, and
hand. Nowadays several research papers still focus on
improving recognition accuracy, although topics such as
interoperability, template protection, scalability, performance
evaluation, and biometric sample synthesis, are receiving much
Performance evaluation of recognition algorithms and
synthesis of biometric data are related to some extent. The
spread of publicly available databases allows researchers to
advance the state-of-the-art. However, there are several barriers
related with the acquisition and dissemination of biometric data
such as: i) the acquisition of large dataset is an expensive and
time consuming task; ii) legal issues for privacy-protection
national laws often do not allow database collection or
distribution; iii) there are many security concerns by users and
organizations. All these issues complicate the spread of
databases among researchers, industry and governments.
Biometric sample synthesis emerged as feasible way to easily
provide large datasets for training and testing algorithms (see
for instance [2]) without any legal or security concern.
The modelling of biometric traits for the generation of
synthetic data is a challenging problem. The model should be
able to generate a large number of samples including realistic
deformation for different poses (intraclass variability) and
feasible differences between subjects (interclass variability). In
Manuscript received ??? ???, ???. This study was funded by Spanish
government MCINN TEC2012-38630-C04-02 research project.
Aythami Morales and Miguel A. Ferrer are with the IDeTIC. University of
Las Palmas de Gran Canaria, Campus de Tafira, E35017 Las Palmas de Gran
Canaria, Spain. (e-mail:,
addition, the generated data should be realistic enough to spoof
state-of-the-art recognition algorithms. As a result, the
generation of synthetic data has attracted the interest of the
scientific community. Models and methods to generate
biometric samples have been recently proposed for various
traits, such as fingerprint [3][4], face [5], iris [6][7][8] and
signature [9][10][11]. Although the performance evaluation of
recognition algorithms is the most popular application of the
biometric synthetic samples, the range of topics is large and
include security evaluation [7][12], duplication of training
samples [10][11], and reverse engineering process [8], among
The hand is very rich in biometric information and it has been
extensively studied by the scientific community. We can
distinguish between hand recognition approaches based on low
resolution sensors [13][14][15] (typically working under 150
dpi) and approaches based on AFIS/APIS devices
[15][16][17][18][19] (typically working at 500 dpi). Acquiring
palmprints at low resolution allow to extract features such as
hand shape, creases and coarse texture, while operating at
higher resolution enables to reliably extract minutiae from both
fingers and palm.
While low-resolution approaches are well suited for many
commercial applications (e.g. access control), high-resolution
is mandatory for forensic analysis in AFIS/APIS, where
minutiae matching remains the leading comparison technique.
While hand biometrics is a well-investigated subject in the
literature, to the best of our knowledge, there are only
preliminary studies about automatic generation of hand-prints.
In particular, [20] deals with generation of low-resolution
palmprints and [21] is a preliminary study on the generation of
high-resolution hand-print images. Although the method
proposed in [21] allows realistic images to be generated, it does
not address the problem of generating multiple impressions of
the same hand and does not include any quantitative
experimentation or validation of the underlying model.
To the best of our knowledge, the only public available high
resolution palmprint database is the THUPALMLAB database
[22] which comprises only 120 palms acquired with a
commercial AFIS/APIS device. In our opinion, the lack of
public available data slows down the state-of-the-art
advancement in this research area.
In this paper we introduce a methodology to generate hand-
print databases with multiple impressions for each hand. The
generated hand-prints simulate high-resolution images captured
at about 500 dpi. The main objective is the synthesis of realistic
Raffaele Cappelli and Davide Maltoni are with the Biometric System
Laboratory, DISI-University of Bologna, Cesena (FC), Italy (e-mail:
{raffaele.cappelli, davide.maltoni}
Copyright (c) 2013 IEEE. Personal use of this material is permitted.
However, permission to use this material for any other purposes must be
obtained from the IEEE by sending a request to
Synthesis and evaluation of high resolution hand-prints
Aythami Morales, Raffaele Cappelli, Miguel A. Ferrer, Davide Maltoni
ridge-line patterns for the generation of large scale databases
which include intra-class and inter-class natural variability. The
generated fingerprints and palm can be used as a whole
biometric trait, or separately. The method proposed is based on
two main steps: the former involves the generation of a master
hand-print, which is basically a complete hand-print image
without noise and with a default positioning. The latter applies
the hand posture model to the master hand-print to derive
multiple impressions that mimic multiple acquisitions of a real
hand. Master hand-prints determine the inter-class variability of
the generated samples, while multiple impressions determine
the intra-class variability. The generation of a whole master
hand-print requires different traits to be integrated in a realistic
way (including creases on phalanges, palm creases,
singularities, etc.); this makes hand-print synthesis more
challenging than the synthesis of other biometric traits such as
fingerprint or iris.
The main contributions of this paper are: i) a complete
methodology to synthesize hand-prints; ii) a ridge-line
orientation generation that takes into account the local influence
of singularities on the palmprint ridge pattern; iii) the
generation of multiple impressions of the same hand by
simulating the natural hand positioning and variations; iv) a
noise model including the most evident sources of perturbations
such as secondary creases, pressure and skin deformation; v)
the design of specific validation experiment to assess to what
extent the proposed models resemble real hand-print features;
vi) a real application to show the usefulness of synthetic hand-
print databases. Additionally, a database with 5000 synthetic
images generated with the proposed model is made publicly
available for further research in biometrics applications.
The rest of this paper is organized as follows: Section 2
introduces the characteristics of the hand-print ridge pattern;
Section 3 describes the generation of master hand-prints;
Section 4 deals with the derivation of multiple hand-prints from
a given master hand-print; validation experiments and
application are reported in Section 5. Finally Section 6 draws
some conclusions.
The hand-print ridge-line pattern is influenced by several
factors, such as creases of the epidermis (primary and
secondary creases) and singularities (loops and deltas), see Fig.
1. From the empirical observation of several hand-prints, it is
evident a correlation between principal creases and ridge-line
orientations on the palm [21]. It is also well-known that the
number and position of the singularities is strongly related to
the ridge-line flow [23]. Other relevant factors are the pressure
made during the transference/acquisition and the anatomical
characteristics of the hand which introduce nonlinear
deformations and missing regions, see Fig.1. The correlation
between the ridge-pattern of the ten fingers and the palm is low
[24][25] and its random nature is crucial to explain its high
The first step of the generation method is the creation of a
master hand-print, which is a complete hand-print image, noise-
free and with normalized posture. The synthesis of hand-prints
Figure 1. Example of AFIS/APIS hand-print image.
Fingerprint singularities and ridge pattern
Palmprint singularities and ridge pattern
Principal creases
Finger creases
Secondary creases
Missing regions due to
absence of contact
Figure 2. Functional schema of the master hand-print generation method
involves the generation of biometric patterns at different levels
and their integration in a consistent way. The proposed
synthesis method assumes that the orientation of the ridge-lines
is determined, at a global level, by the principal creases, and, at
a local level, by the number and location of the singularities.
Fig. 2 shows a functional schema of the proposed method: the
contour of the hand-print is first generated and then used to
guide subsequent steps; creases and singularities are separately
generated for palm and fingers and then exploited to generate
the orientation image of the whole hand-print; finally the
orientation image together with local frequency information
drives the iterative ridge-line pattern growing. The following
subsections detail the main building blocks.
A. Hand-shape
Hand-shape synthesis is based on the Active Shape Model
(ASM) [12][26]. Hand contours obtained from the GPDS
database [27] are used as ASM training set (a single hand image
from 150 subjects); 14 landmarks from each contour are
automatically located using the methodology proposed in [28].
The hand contour is represented as a  element vector
composed by the  coordinates of    selected points,
uniformly distributed between the landmarks. Principal
Component Analysis (PCA) is applied to determine the main
modes of variation of the hands; the hand contour of a master
hand-print can be then generated as:
      
where is mean contour, is the PCA projection matrix ( 
), whose columns are the eigenvectors of the covariance
matrix, and    is a vector with randomly
generated shape parameters. In our experiments we used  
 modes of variation, which covers 99.9% of the trained set
variance. To obtain realistic hand-shape images, each shape
parameter is randomly generated in the range  
, being the  eigenvalue of the covariance matrix.
The hand-shape delimits the ridge-line pattern and the size of
the hand. In AFIS/APIS devices the hand-shape is not stable
due to the distortions introduced by the pressure and the optical
characteristics of the devices, see Fig. 1. Such nonlinear
distortions will be included in subsequent generation steps.
Figure 3. Skeleton of the generated hand-shape (dashed line) and creases image
(continuous line) with palm lines (life, heart and head) and creases which
divide the finger phalanges (distal, medial and proximal).
B. Creases
A pixel-wise binary crease image is synthesized by
combining palm and finger creases (Fig. 3), which are
generated as described in the following paragraphs.
Palm creases
The number, position and shape of palm creases are strongly
related to the joints and muscles under the skin surface. In
nature there are many possible configurations of three main
palm creases: the radial longitudinal transverse crease, distal
transverse crease and proximal transverse crease.
Regarding the presence of the principal creases and the
number of the intersections between them, [29] proposed a
classification into six categories. In this work we consider the
two most common categories which can be found in 90% of the
people: category 4 (palmprints composed of all three principal
creases and no intersection) and category 5 (palmprints
composed of all three principal creases and one intersection).
The category of a given master hand-print is randomly selected
(with 85% and 15% probability, according to [29]). The main
difference between the two categories is the starting point of the
major hand creases. To derive a model for the shape and
position of true palm creases, we manually marked the principal
creases of 150 palms taken from the GPDS database (12 points
per line). Then each principal crease is modeled by means of an
ASM with 10 modes of variation, with an approach analogous
to that described in section III-A. The generation of the creases
is carried out as follows: 1) the starting and ending points are
positioned within bounding boxes whose placement depends on
the hand-shape landmarks; 2) the shape of each of the three
creases is synthesized using the ASM model; 3) random offsets
are applied to the starting and ending point in order to introduce
more variability.
While the aspect and position of the primary palm creases
have been largely studied [29], there are not many studies on
the nature and formation of secondary or minor creases. The
discrimination capability of the secondary creases in low-
resolution palmprint recognition systems is well recognized
[13][14][15]. However there are difficulties to obtain a stable
secondary creases pattern in AFIS/APIS systems [16][17]. In
particular, some secondary creases may be absent or non-visible
Ridge-line pattern
Creases generation
Orientation image
Singularity generation
Hand shape
Frequency image
Radial crease
Proximal crease
based on the skin state and amount of pressure. For this reason
we prefer to define number, position and sizes of the secondary
creases for a master hand-print but to add them (with a given
probability) only during the generation of multiple impression
(see Section IV.B).
Finger creases
Human hands contain fourteen digital bones, also called
phalanges. Excluding the thumb, each finger contains three
phalanges, which, starting from the fingertip are: distal, medial
(not present in thumbs), and proximal. The skin over the region
between two phalanges is typically characterized by some
horizontal creases due to the flexion of joints.
Creases on phalanges are here generated by simply dividing
the finger into two or three parts with horizontal lines. In the
proposed generation method, the exact placement of the creases
is determined according to the mean distances between
phalanges as reported in [30] (see Table I), with the addition of
random offsets of ±5% to introduce inter-class variability.
Some salt and pepper noise is finally added to the creases to
make them more realistic (Fig. 3).
C. Singularities
A set of singularities (loop and delta points) is randomly
generated for the palm and the fingers, as described in the
following paragraphs.
Palm singularities
The presence of loop and delta singularities is locally related to
the orientation image of palm and fingers. The distribution of
the palm singularities (number, position and type) has not been
much investigated in the literature. In the case of fingerprint
singularities, some researchers observed that “for each
completely captured fingerprint, there are the same numbers of
loops and deltas” [31]. This conclusion is based on the
assumption that if a fingerprint is captured completely, the left,
right, bottom, and top boundaries are nearly horizontal. This
helps to model fingerprint singularities, but does not seem to
generalize to palms, because the palm ridge pattern shows a
mixture of horizontal, vertical and diagonal orientations at its
boundaries. The singularities of the fingerprint is determined by
a small number of categories, five in the most widely adopted
classification: Arch, Tented arch, Right loop, Left loop and
Whorl [24].
In order to realistically generate palm singularities, we
inferred some heuristic rules from a visual analysis of the
THUPalmLab database (500 dpi palmprints) [22] and the
results of the study in [23] (note that most existing palmprint
databases, such as the GPDS database [27], cannot be used to
this purpose because of the lack of ridge-flow information due
to the low image resolution). In particular, we assume that, for
a generic palm, there is:
a delta singularity near the base of each finger (except for
the thumb).
80% probability of finding a delta singularity near the
bottom region of the palm (carpal delta).
60% probability of finding a loop singularity between two
finger’s deltas or near the palm boundaries. Therefore the
number of the loops is in the range [0-4].
The proposed generation method chooses the number and
position of palm singularities according to the above
assumptions: the total number of singularities may vary
between four and nine: each location is randomly chosen
according to the hand-shape and a 2D Gaussian distribution.
Fig. 4 shows examples of singularities superimposed to the
creases image .
Finger singularities (distal phalange)
For each finger, the finger class and the singularities are
generated according to the probability distributions introduced
in [32].
D. Orientation image
The orientation image is a matrix whose elements encode the
local ridge-line orientations [33]. A block-wise (  pixels)
palmprint orientation image encodes, for each block of center
 (represented as a complex number     ), an
angle   .
is computed as the sum of two orientation images: 
(obtained from palm and finger creases) and  (from palm and
finger singularities):
    
The orientation image  is obtained by finding the principal
axis of variation in the image gradients [34] extracted from the
creases image :
  
 
 
 
   
 
where corresponds to 
, the partial derivate in
(horizontal) direction, and corresponds to 
, the
partial derivate in (vertical) direction on a local window of
size   pixels. The orientation at each point is computed
as 
   
. Fig. 4 shows an example of
resulting image.
The orientation image  is generated by a weighted version
(proposed in this paper) of the Sherlock and Monro model [35],
where the influence of each singularity rapidly vanishes moving
far from its location.
 
where  is computed with modulus π, and:
    are the coordinates of the loop
    are the coordinates of the delta
    and are the phase angle and
modulus of the complex number , respectively;
is a parameter that controls the area of influence of each
singularity: in our experiments we have randomly selected
a value in the range [0.015-0.045] for each singularity.
Fig. 4 shows the results obtained using both the original
Sherlock and Monro model and the weighted version proposed
in this paper.
D. Ridge-line pattern
To generate a realistic ridge-line pattern, the method proposed
in [3] for fingerprints, is here adopted to the whole hand-print.
Given the orientation image and a block-wise frequency
image  as input, the method consists in iteratively enhancing
a white image containing just a few random black pixels,
through Gabor contextual filters tuned according to the local
orientations and frequencies. The frequency image  is here
simply generated using a constant frequency over the whole
pattern, except near principal creases, where the local frequency
is set to a much smaller value; some random perturbations and
smoothing are finally applied to increase the pattern variability.
A further post-processing is carried out on the ridge-line
pattern image to highlight the creases: this operation simply
consists in a saturated pixel-wise addition of the creases image
. Fig. 5 shows an example of master hand-print resulting from
the proposed approach.
The most important sources of variability in impressions of the
same hand are given by: hand positioning, finger spread, and
pressure against the scanner surface. Simulating different hand
positioning (in terms of translation and rotation) is
straightforward. Variation of finger spread, secondary creases
and pressure pattern are dealt with in the following subsections.
Figure 4. From left to right and top to bottom: a) Principal creases orientation
image ; b) Sherlock and Monro orientation image (triangles for deltas and
crosses for loops); c) Proposed Sherlock and Monro model with sigmoid
correction ; d) Palmprint orientation image  obtained combining the
principal creases orientation  and the singularity orientation image
A. Finger spread
In order to simulate varying finger spreads, the probability
distributions of finger positioning (angles) have been estimated
Figure 5. An example of master hand-print generated by the proposed methodology.
from GPDS database [27]; for each hand-print to be generated,
rotation angles for the five fingers are drawn from the
corresponding distributions. A new hand contour is then
obtained by rotating the control points of each finger according
to its angle (Fig. 6). Points in the left and right palm boundaries
are also randomly perturbed to simulate small variations of the
palm width.
At this stage, the master hand-print image has to be adjusted
to the hand-shape generated for a given impression. It is worth
noting that the finger movement produces a certain amount of
skin distortion in some regions of the palm, in particular for the
thumb. In order to simulate such skin distortion, the following
image warping technique is applied to the master hand-print:
1. A Delaunay triangulation [36] is calculated over 160
control points of the master hand-print shape (30 equally
spaced points for each finger and ten points for the rest of
the palm boundary, sampled according the landmarks of
the ASM model presented in section III.A);
2. The corresponding control points on the hand-shape of the
impression to be generated allow to map each triangle of
the master hand-print to the corresponding triangle in the
3. The position of each pixel  in the impression is
mapped to a position  in the master hand-print
(reverse mapping), by considering the relative position of
 inside the triangle that it belongs to (this can be
done using barycentric coordinates);
4. The value of each pixel  is determined by resampling
the master hand-print at the corresponding position 
using a simple interpolation technique.
Fig. 7 shows an example of the above technique and
highlights the skin distortion which is produced according to
the finger movements.
B. Pressure pattern, secondary creases and noising
The final aspect of any hand-print image depends on several
factors, including the shape of the hand, the acquisition device,
and the way the user applies pressure against the acquisition
surface. Real hand-print images include regions where the
pattern is totally missing or scarcely visible and regions where
the pattern is visible but the quality is very low due to the
presence of noise.
The proposed approach simulates the effects of both non-
uniform pressure and other sources of noise and incorporates
them into a single transform that is applied to the binary ridge-
line pattern, resulting in a realistic gray-level hand-print.
The simulation of non-uniform pressure is dealt with as
follows. A set of 23 hand-print images from 11 different hands
have been acquired with varying pressure in both weight and
direction against the sensor surface. The major differences are
due to the amount of pressure made by the subject and the
anatomic characteristics of the hand. The 23 templates were
chosen among a larger set because they characterize most
pressure profiles. The contour of each hand-print has been
manually labeled and normalized to match the mean contour
introduced in section III.A. A pressure template is then derived
from the normalized hand-print, by estimating the “pressure”
level of each pixel according to its gray-level, after applying
median filtering to the image (Fig. 8).
Figure 7. Triangulation-based image warping applied to the control points on
the master hand-print to adjust the ridge-line pattern to the shape of a given
impression. a) Triangulation of the master hand-print; b) Triangulation of the
impression; c) Triangulation superimposed to the master-hand-print ridge-line
pattern; d) Result of the image warping; e) and f) portions of the two ridge-
line patterns superimposed to highlight the skin distortion.
Figure 6. Hand-shape of a master hand-print (dotted line) and of a given
impression (continuous line).
The 23 pressure templates include variations such as portions
of fingers and palm missing and heterogeneous distribution of
pressure among different regions. For each master hand-print, a
corresponding master pressure-image  is generated by
combining three randomly-chosen templates, using a weighted
sum with weights sampled from a uniform distribution (sum of
weights equals to 1). For each hand-print to be generated, the
corresponding pressure image is obtained by carrying out the
following steps on :
1. Applying the same transformations made on the master
hand-print to simulate hand positioning and finger spread
(see section IV.A), so that the pressure image is perfectly
aligned with the hand-print ridge-line pattern;
2. The % pixels with the lowest pressure level in are set
to zero (no-pressure), where  
 
 and
are values sampled from two Pareto distributions [37] with
scale parameter 0.3 and 0.05, respectively, and shape
parameter 0.1.
is sampled once for each master hand-
print and controls the average pressure-level of the master-
handprint (inter-class variability), while
 is sampled for
each hand-print to be generated and controls the
characteristics of a specific impression (intra-class
variability). This probability distribution is chosen
because of its effectiveness in describing natural
observable phenomena.
3. Finally histogram equalization is performed on to
obtain the final pressure image (Fig. 9).
The pressure image is used in the following procedure to
simulate artifacts caused by uneven pressure, minor creases and
other noise sources:
1. The secondary or minor creases of the hand show a
highly random nature. The size, position and orientation
Figure 8. A pressure template: a) Original hand-print image; b) Resulting
pressure template. Light gray values denote low pressure (white
Figure 10. An example of final hand-print generated by the proposed method.
Secondary creases
Primary crease
Fingerprint singularities and ridge pattern
Palmprint singularities and ridge pattern
Loop and Delta
singular points
Figure 9. Examples of pressure-images: a) Generated pressure image with
; b)
; c)
. Light gray values denote low
pressure (white no-pressure).
vary for each person and also for each region of the hand.
For each master hand-print, the minor creases are
simulated as ellipses with a high eccentricity, random
size, random angle and random position inside the hand.
The number of secondary creases is chosen randomly in
the range 40-90. Each ellipse is curved using a cosine
function with variable frequency to simulate certain
level of curvature in the creases. Once established the
number, position and sizes of the secondary creases for
the master hand-print, the creases are added to the
multiple impressions with a 10% miss-probability;
2. Valley white pixels in the ridge-line pattern are copied into
a separate layer;
3. For each pixel , a small white blob (with random size
between    and    pixels) is applied to the ridge-
line pattern with probability  
, where is a noise probability which is
progressively higher towards the contours of the hand;
4. The resulting image is smoothed using a    Gaussian
5. The valley layer is superimposed to obtain the final result.
Fig. 10 shows an example of realistic grayscale hand-print
image obtained with the proposed approach.
The validation of the synthetic data is based on: a) quantitative
comparison between synthetic and real data; and b) its
usefulness in a real application.
The number of features that can be extracted from only one
synthetic hand-print is very large. For that reason, it is well
accepted [3][4][6][7][8][9] the validation of biometric synthetic
data based on the study of its performance using state-of-the-art
recognition algorithms. The proposed generation method
focuses on obtaining a realistic ridge-line pattern. Because
minutiae are the basis of AFIS/APIS systems [16][17][19][33],
the synthetic images will be evaluated using a state-of-the-art
minutiae extractor and matcher.
A    hand-print database (1000 hands with five
impressions each) has been generated with the proposed
methodology and used in the experiments reported in the
following sections. This database is available at
A. Validation of the hand-print minutiae
In order to validate the generated ridge-line patterns at the
minutiae-level, the match score distributions obtained from the
synthetized hand-print database have been compared to those
from real images using the Minutia Cylinder-Code (MCC)
matching algorithm [38][39] and the state-of-the-art minutiae
extractor described in [17]. The MCC descriptor is a minutiae
representation based on 3D binary structures built from
minutiae positions and angles. Each cylinder contains
information about the neighborhood of each minutia and this
information is encode using a fixed-length, and bit-oriented
algorithm. The performance achieved by MCC in international
competitions is one of the most competitive among all the
academic approaches [39][40]. Since, to the best of our
knowledge, no high-resolution (at least 500 dpi are required for
minutiae extraction) hand-print databases are publicly
available, the THUPalmLab palmprint database [22] has been
used to generate match score distributions from real samples.
Fig. 11 shows the genuine and impostor score distributions
obtained in this validation experiment. It can be observed that
the score distributions for the real and synthetized database
exhibit a similar trend. This result suggests that minutiae
generated in the synthetic hand-prints, as well as related
features (such as local ridge-line orientations, which are critical
during minutiae extraction), are realistic, at least from the point
of view of the matching algorithm adopted.
Figure 11. Match score distributions of the MCC algorithm on the synthetic and
real databases. Scores have been normalized between [0,1].
B. Optimizing the parameters of a minutiae-based matcher
The synthetic hand-print images generated can be used in
variety of applications. The next section proposes an
optimization application to illustrate the usefulness of synthetic
hand-print databases.
An interesting application of the proposed hand-print
synthesis method is to provide large amount of data for
optimizing the parameters of biometric matching algorithms.
This section describes an experiment carried out using the MCC
algorithm [38]. Since MCC was originally developed for
fingerprint recognition, its parameters have been originally
optimized on a large fingerprint database in [39]. In order to
successfully use MCC for palmprint or hand-print matching, it
is necessary to properly adjust its parameters. Obviously,
parameter adjustment cannot be done on the final test database,
to avoid data over-fitting: the availability of realistic synthetic
data is then very useful in this context. The main MCC
parameters have been tuned on the    synthetic hand-
print database mentioned in the previous section, by performing
an exhaustive search over a reasonable set of parameter values.
Table II reports the parameters (the reader should refer to [38]
for a description of their meaning) with the corresponding
values, as modified after the above described tuning procedure.
The MCC algorithm with optimized parameters has been
tested on the PV-FULL-1.0 benchmark on FVC-onGoing [40],
which contains images of palms acquired in operational
conditions using high-quality optical scanners. Table III reports
the EER obtained using MCC native parameters (as suggested
in [39]) and the new parameters tuned on the synthetic dataset;
the table also reports results of other state-of-the-art algorithms
published on FVC-onGoing.
Native value for
fingerprints (see in [39])
optimized value
3, 10
10, 20
, 
, 
, 
, 
 , 
 , 
RegionGrow+GlobalMatch (Department of Automation, Tsinghua
MinutiaeClusterFull (Dept. of Computer Science & Engineering,
Michigan State University)
Mnt (Institute of Automation, Chinese Academy of Sciences)
MCC with original parameters in [39]
MCC optimized using the synthetic hand-print dataset
This paper introduces a novel method for the generation of
realistic high-resolution synthetic hand-prints similar to those
acquired by AFIS/APIS devices (see Fig. 12). Specific traits
such as fingerprint, palmprint (at minutiae level) and hand-
shape have been analyzed and modeled in their main degrees of
freedom. Other traits such as the creases of the palm and
phalanges have been introduced to increase the realism.
The effectiveness of the synthetic hand-print images for
testing purposes has been assessed by comparing results on real
and synthetic databases. We have also demonstrated the
usefulness of synthetic hand-print samples for system
optimization. We believe that besides the optimization here
proposed the availability of large dataset of hand-prints can be
useful for other emerging approaches such as dictionary
learning (see [41][42]).
Some of the sub-models here proposed have been derived by
quite small datasets of real samples (often manually annotated),
possibly resulting in over-simplifications. In the future we
intend to study, more in detail, the feature variation (intra-class
and inter-class) of each specific feature-type, in order to
develop more accurate models including all biometric traits on
the hand.
Finally some notes on computation complexity: the proposed
generation method can generate a database of one thousand
synthetic identities in about 12 hours. The generation time
varies acording the characteristic of the generated palmprints
and Table IV shows the approximate time requeriments to
generate one synthetic identity using a Intel i7-860 Processor
(8M Cache, 2.80 GHz, 8Gb RAM) with a single-thread
implementation (i.e., not exploiting multiple CPU cores).
Time (sec)
Master hand-print Generation
Impression Generation (five samples)
[1] A.K. Jain, R. Bolle, and S. Pankanti, Biometrics, Personal Identification
in Networked Society, Kluwer Academic Publishers, 2001.
[2] R. Cappelli, M. Ferrara, A. Franco, D. Maltoni, “Fingerprint verification
competition 2006”, Biometric Technology Today, vol. 15, no. 7-8, pp 7
9, Jul.-Aug. 2007.
[3] R. Cappelli, Synthetic Fingerprint Generation, In D. Maltoni, D. Maio, A.
K. Jain, and S. Prabhakar, Eds., Handbook of Fingerprint Recognition, pp.
203-232, Springer, 2009.
[4] Qijun Zhao, A. K. Jain, N. G. Paulter, M. Taylor, "Fingerprint image
synthesis based on statistical feature models", in Proc. IEEE Fifth Int.
Conf. on Biometrics: Theory, Applications and Systems (BTAS), pp. 23-
30, 2012.
[5] Blanz and T. Vetter, “A Morphable Model for the Synthesis of 3D Faces”,
in Proc. of the 26th annual conf. on Computer graphics and interactive
techniques, pp. 187-194, 1999.
[6] A. Lefohn, B. Budge, P. Shirley, R. Caruso, and E. Reinhard, “An
Ocularists Approach to Human Iris Synthesis”. IEEE Computer Graphics
and Applications, vol. 23, issue 6, pp. 70-75, Nov. 2003.
[7] S. Venugopalan, M. Savvides, How to generate spoofed irises from an
iris code template”, IEEE Trans. on Information Forensics and Security,
vol. 6, pp. 385394, Jun. 2011.
[8] Luís Cardoso, André Barbosa, Frutuoso Silva, António M. G. Pinheiro,
Member, IEEE, and Hugo Proença, “Iris Biometrics: Synthesis of
Degraded Ocular Images”, IEEE Trans. on Information Forensics and
Security, vol. 8, no. 7, pp. 1115-1125, Jul. 2013.
[9] J.Galbally, R. Plamondon, J. Fierrez and J. Ortega-Garcia, "Synthetic on-
line signature generation. Part I: Methodology and algorithms", Pattern
Recognition, vol. 45, pp. 2610-2621, July 2012.
[10] J. Galbally, J. Fierrez, M. Martinez-Diaz, J. Ortega-Garcia, “Improving
the enrollment in dynamic signature verification with synthetic samples”,
Proc. of the IAPR Int. Conf. on Document Analysis and Recognition, pp.
1295-1299, Jul. 2009.
[11] C. Rabasse, M. Guest, C. Fairhurst, “A New Method for the Synthesis of
Signature Data With Natural Variability”, IEEE Trans. on System, Man,
and Cybernetics Part B, vol. 38, no. 3, Jun. 2008.
[12] Marta Gomez-Barrero, Javier Galbally, Aythami Morales, Miguel A.
Ferrer, Julian Fierrez, Javier Ortega-Garcia, "Inverse Biometrics: A Case
Study in Hand Geometry Authentication", in Proc. of 21st Int. Conf.
on Pattern Recognition, pp. 1281-1284, 2012.
[13] D. Zhang, 2004. Palmprint Authentication. Kluwer Academic.
[14] D. Zhang, W-K. Kong, J. You, and M. Wong, Online Palmprint
Identification”, IEEE Trans. on Pattern Analysis Machine Intelligence,
vol. 20, no. 10, pp. 17411756, 2008.
[15] A. Kong, D. Zhang, M. Kamel, A survey of palmprint recognition”,
Pattern Recognition, vol. 42, no. 11, pp. 2841-2849, 2009.
[16] A. K. Jain, F. Jianjiang, Latent Palmprint Matching”, IEEE Trans. on
Pattern Analysis and Machine Intelligence, vol. 31, pp. 1032-1047, 2009.
[17] R. Cappelli, M. Ferrara, D. Maio, A Fast and Accurate Palmprint
Recognition System based on Minutiae, IEEE Trans. on Systems, Man
and Cybernetics - Part B, vol. 423, pp. 956-962, 2012.
[18] X. Yang, J. Feng, J. Zhou, Palm print Indexing Based on Ridge
Features”, Proc. of IEEE Int. Joint Conf. on Biometrics, Washington DC,
USA, pp. 1-8, 2011.
[19] A. K. Jain and J. Feng, “Latent Fingerprint Matching,” IEEE Trans. on
Pattern Analysis Machine Intelligence, vol. 33, no. 1, pp. 88100, 2011.
[20] Wei Zhuoshi, Yufei Han, Zhenan Sun, Tieniu Tan, “Palmprint image
synthesis: A preliminary study”, in Proc. of 15th IEEE Int. Conf. on Image
Processing, pp. 285-288, 2008.
[21] J. Feng, H. Zhou, and Jie Zhou, “A Preliminary Study of Hand-print
Synthesis”, in Proc. of Chinese Conf. on Biometric Recognition, 2011,
[22] Jifeng Dai, Jie Zhou, "Multi-feature Based High-resolution Palmprint
Recognition", IEEE Trans. on Pattern Analysis and Machine Intelligence,
vol. 33, no. 5, pp. 945-959, May 2011.
[23] Yan Zheng, GuangShun Shi, QingRen Wang, "Location of Special Areas
on Palmprint”, in Proc. of Int. Cong. on Image and Signal Processing,
2008, pp.786-791.
[24] E.R. Henry, Classification and Uses of Finger Prints. Routledge, 1900.
[25] Chatterjee, S. K. Finger, Palm and Sole Prints; Artine Press: Calcutta,
India, 1953.
[26] T.F.Cootes, C.J.Taylor, D.H.Cooper and J.Graham, Active Shape
Models Their Training and Application”. Computer Vision and Image
Understanding, vol. 61, no. 1, pp. 38-59, January 1995.
[27] M. A. Ferrer, A. Morales, C. M. Travieso, J. B. Alonso, “Low Cost
Multimodal Biometric identification System Based on Hand Geometry,
Palm and Finger Print Texture”, in Proc. of 41st Annual IEEE Int.
Carnahan Conf. on Security Technology, 2007, pp. 52 58.
[28] Miguel A. Ferrer and Aythami Morales. Hand-Shape Biometrics
combining the visible and Short Wave InfraRed Bands”, IEEE Trans.
Information Forensics & Security, vol. 6, no. 4, pp. 1305-1314, Dec.
[29] Xiangqian Wu, David Zhang; Kuanquan Wang, Bo Huang, “Palmprint
classication using principal lines”, Pattern Recognition, vol. 37, no. 10,
pp. 19871998, Oct. 2004.
[30] T. Greiner, Hand anthropometry of U.S. Army personnel, U.S. Army
Natick Res., Development and Eng. Center, Dec. 1991.
[31] J. Zhou, J. Gu and David Zhang, “Singular Points Analysis in Fingerprints
based on Topological Structure and Orientation Field”, in Proc. of Int.
Conf. on Biometrics, pp. 261-270, 2007.
[32] R. Cappelli, D. Maltoni, “On the Spatial Distribution of Fingerprint
Singularities”, IEEE Trans. on Pattern Analysis and Machine
Intelligence, vol. 31, no. 4, pp. 742748, Apr. 2009.
[33] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, Handbook of
Fingerprint Recognition. New York: Springer-Verlag, 2009.
[34] Peter Kovesi , MATLAB and Octave Functions for Computer Vision and
Image Processing, School of Computer Science & Software Engineering,
The University of Western Australia, 2005, available online:
[35] B.G. Sherlock, D.M. Monro, “A model for interpreting fingerprint
topology”, Pattern Recognition, vol. 26, no. 7, pp. 10471055, Jul. 1993.
[36] Mark de Berg, Otfried Cheong, Marc van Kreveld, Mark Overmars,
Computational Geometry: Algorithms and Applications. Springer-Verlag,
[37] H. Cramér, Mathematical methods of statistics, Princeton Univ. Press,
[38] R. Cappelli, M. Ferrara, D. Maltoni, “Minutia Cylinder-Code: a new
representation and matching technique for fingerprint recognition”, IEEE
Trans. on Pattern Analysis Machine Intelligence, vol. 32, 21282141,
Dec. 2010.
[39] R. Cappelli, M. Ferrara, D. Maltoni and M. Tistarelli, "MCC: a Baseline
Algorithm for Fingerprint Verification in FVC-onGoing", in Proc. of 11th
Int. Conf. on Control, Automation, Robotics and Vision, 2010, pp. 19-23.
[40] (22/01/2014). [Online]. FVC-onGoing web site:, March 2013.
[41] J. Feng, J. Zhou, A.K. Jain, "Orientation field estimation for latent
fingerprint enhancement", IEEE Trans. on Pattern Analysis Machine
Intelligence, vol. 35, no. 4, pp. 925-940, 2013.
[42] X Yang, J Feng, J Zhou, “Localized Dictionaries Based Orientation Field
Estimation for Latent Fingerprints”, IEEE Trans. on Pattern Analysis
Machine Intelligence, vol. 36, no. 5, pp. 955969, 2014.
Aythami Morales received his M.Sc. degree in
Telecommunication Engineering in 2006 from
Universidad de Las Palmas de Gran Canaria. He
received his Ph.D degree from La Universidad de Las
Palmas de Gran Canaria in 2011. He performs his
research works in the ATVS Biometric Research
Lab, Universidad Autónoma de Madrid and he has
undertaken research visits to the Biometric Research
Laboratory at Michigan State University, the Biometric
Research Center at Hong Kong Polytechnic University
and the Biometric System Laboratory at University of Bologna. His research
interests are focused on pattern recognition, computer vision, machine learning and
biometrics signal processing. He is the author of more than 30 scientific articles
published in international journals and conferences.
Figure 12. Examples of: a) generated synthetic hand-print and b) real hand-print.
Raffaele Cappelli (M’09) received the Laurea
degree cum laude in computer science from the
University of Bologna, Cesena, Italy, in 1998. In
2002, he received the Ph.D. degree in computer
science and electronic engineering at DEIS,
University of Bologna. He is an Associate
Researcher at the University of Bologna, Italy. He
teaches “Pattern Recognition” at Computer
Science, and he is a member of the Biometric
System Laboratory, University of Bologna,
Cesena, Italy. His research interests include pattern recognition, image
retrieval by similarity, and biometric systems (fingerprint classification
and recognition, synthetic fingerprint generation, fingerprint aliveness
detection, fingerprint scanner quality, face recognition, and performance
evaluation methodologies).
Miguel A. Ferrer received the M.Sc. degree in
telecommunications, in 1988, and his Ph.D. degree, in
1994, each from the Universidad Policnica de Madrid,
Spain. He belongs to the Digital Signal Processing
research group (GPDS) of the research institute for
technological development and Communication
Innovation (IDeTIC) at the University of Las Palmas de
Gran Canaria in Spain where since 1990 he has been an
Associate Professor. His research interests lies in the fields
of computer vision, pattern recognition, biometrics,
mainly those based on hand and handwriting, audio quality, mainly for health and
condition machinery evaluation and vision applications to fisheries and aquaculture.
Davide Maltoni (M’05) is an Associate
Professor at the Department of Electronics,
Informatics and Systems (DEIS), University of
Bologna. He teaches “Computer Architectures”
and “Pattern Recognition” in Computer Science,
University of Bologna, Cesena. His research
interests are in the area of Pattern Recognition
and Computer Vision. He is active in the field of
Biometric Systems (fingerprint recognition, face
recognition, hand recognition, performance
evaluation of biometric systems). He is codirector of the Biometric
Systems Laboratory, Cesena, Italy, which is internationally known for its
research and publications in the field. He is author of two books:
Biometric Systems, Technology, Design and Performance Evaluation
(Springer, 2005) and Handbook of Fingerprint Recognition (Springer,
2003; II edition 2009), for which received the PSP award from the
Association of American Publishers.
... Synthetic data is generated from a population model, and used to test datasets, to validate mathematical models and to train machine learning algorithms. One problem is 'How well the synthetic data replicates the authentic data?' [31], [32], [51]. In this paper, we formulate it as follows 'How risky this replacement?' and 'Can we trust this synthetic 'life' attributes?' Partial answers can be found in [5]. ...
... 1) Synthetic biometrics such as face [49], handprints [32], speech [39], signatures [15], iris [7], [27] for testing bio-metric algorithms [33] and modeling critical scenarios such as biometric attacks [27]; 2) Other synthetic data, e.g. for the sensors that detect concealed (illicit) items; it is usually radar illumination to detect knives, pistols, grenades) [22], [43]; 3) AI decision assistance such as avatar-like humanmachine interfaces [52]; this concept is analogue to the engineered life form concept [1], [5]. ...
... Impact of synthetic data on performance and privacy in complex dynamical systems such as security checkpoint is a challenging problem. Various aspects of this problem were studied in [7], [32], [54], [55]. Privacy issues of synthetic data were discussed in [5]. ...
... Synthetic hand prints: The hand print is a hybrid highresolution biometric trait that combines fingerprints, the palmprint, and hand-shape. A synthesizer of hand prints is proposed in [116]. This is an example of a synthetic multi-biometric. ...
Full-text available
This paper aims at identifying emerging computational intelligence trends for the design and modeling of complex biometric-enabled infrastructure and systems. Biometric-enabled systems are evolving towards deep learning and deep inference using the principles of adaptive computing, - the front tides of the modern computational intelligence domain. Therefore, we focus on intelligent inference engines widely deployed in biometrics. Computational intelligence applications that cover a wide spectrum of biometric tasks using physiological and behavioral traits are chosen for illustration. We highlight the technology gaps that must be addressed in future generations of biometric systems. The reported approaches and results primarily address the researchers who work towards developing the next generation of intelligent biometric-enabled systems.
... The main aim of the generation method proposed there was to reconstruct hand images from the templates, so it is a special case of synthesis (new identities are not generated). Finally, handshape synthesis (Feng et al., 2011;Morales et al., 2014) has been integrated as part of hand-print generation focused on ridge-line patterns. The aim of the hand-shape synthesizer in the previous work was the generation of realistic hand-shape contours (in terms of appearance exclusively) to be integrated in the hand-print model. ...
This work proposes and analyzes a novel methodology for hand-shape image synthesis. The hand-shape is a popular biometric trait with a high convenience of use and non-intrusive acquisition. The proposed algorithm allows to generate realistic images with natural intra-person and inter-person variability. The method is based on the Active Shape Model algorithm which has been modified in order to add the biometric information typical of new synthetic identities. The generated images are evaluated using three public databases and two hand-shape recognition systems. The results show the suitability of the synthetic data for biometric recognition works. In addition, two novel applications have been proposed to provide new insights in hand-shape biometric recognition including: improvement of machine learning classification based on synthetic training sets and scalability analysis of hand-shape biometrics when the population of the database is increased by two orders of magnitude with respect to existing databases.
Conference Paper
Full-text available
Ebben a cikkben a biometrikus beléptető rendszerek mindennapi üzemeltetésnél felmerülő felhasználói együttműködés szükségessége kerül megvizsgálásra. Minden biometrikus berendezés megfelelő működéséhez feltétel az azonosítandó biometrikus minta optimális pozícióban történő prezentálása. A nemzetközi és hazai szakirodalom ezt a tényezőt külön nem vizsgálja, azonban a gyakorlati tapasztalatok és tesztek alapján elmondható, hogy egy biometrikus technológia vagy rendszer áteresztőképességét jelentősen képes befolyásolni a minta megfelelő pozicionálása. A kontaktusmentes technológiák teljes szabadságot nyújtanak a felhasználónak a minta megfelelő elhelyezésére, míg egyes kontaktusos technológiák fizikailag rögzítik a helyes pozíciót. A gyakorlati tapasztalatok azt mutatják, hogy a rögzített helyzet mindig jobb felismerési pontosságot eredményez, azonban jelentősen magasabb a felhasználók ellenérzése. A biometrikus eszközök pozicionálási tesztek célja az eszközök vizsgálata olyan szempontból, hogy azok mennyire érzékenyek a minta megfelelő elhelyezésére. A pozícionálás nem csak az eszközök kialakításától, hanem a felhasználók ügyességétől, képzettségétől és hajlandóságuktól is függ. A tesztek végeredménye alapján felmérhető, hogy az azonosítási eljárás során milyen mértékű együttműködést igényel az eszköz a felhasználó részéről és a használat során mennyire elvárható a sikeres azonosítás. A pozicionálási tesztek magukba foglalják a minta forgatását és eltolását valamint a szenzortávolság változtatását. A biometrikus minta kontrollált mozgatásával meghatározható az eszköz hibás elutasítás arányának függése az optimális pozicionálástól eltérő esetekben és ez alapján következtetni lehet a várható teljesítményére. Meghatározásra kerül egy berendezés szabadságfokainak és kötöttségeinek száma, valamint a biometrikus azonosítás automatizálhatósága.
Conference Paper
Full-text available
Fingerprint image synthesis has received considerable attention because of its potential use in generating large databases to evaluate the performance of fingerprint recognition systems. Existing fingerprint synthesis algorithms (e.g., SFinGe) focus on rendering realistic fingerprint images, but the features (e.g., minutiae) in these fingerprints are formed in an uncontrollable manner. However, generating synthetic fingerprint images with specified features is more useful in developing, evaluating and optimizing fingerprint recognition systems by providing ground truth features in the synthesized images. In this paper, we propose a method to synthesize fingerprint images that retain prespecified features (i.e., singular points, orientation field, and minutiae). To obtain realistic fingerprints, these features are sampled from appropriate statistical models which are trained by using real fingerprints in public domain databases. We validate the proposed method by comparing the synthesized images with those generated by SFinGe and by investigating the match score distributions on synthesized and real fingerprint databases. Furthermore, the synthesized fingerprint images and their minutiae are used to evaluate the matching capabilities of two commercial off-the-shelf (COTS) fingerprint matchers.
This paper describes SFinGe, a method for generating synthetic fingerprints on the basis of some mathematical models that describe the main features of real fingerprints. The synthetic images are randomly generated according to few given parameters. SFinGe captures the variability which characterizes the acquisition of fingerprints through on-line sensors and uses a sequence of steps to derive a series of "impressions" of the same "artificial finger". The approach is able to generate very realistic fingerprints, which can be useful for performance evaluation, training and testing of fingerprint-based systems.
Dictionary based orientation field estimation approach has shown promising performance for latent fingerprints. In this paper, we seek to exploit stronger prior knowledge of fingerprints in order to further improve the performance. Realizing that ridge orientations at different locations of fingerprints have different characteristics, we propose a localized dictionaries-based orientation field estimation algorithm, in which noisy orientation patch at a location output by a local estimation approach is replaced by real orientation patch in the local dictionary at the same location. The precondition of applying localized dictionaries is that the pose of the latent fingerprint needs to be estimated. We propose a Hough transform-based fingerprint pose estimation algorithm, in which the predictions about fingerprint pose made by all orientation patches in the latent fingerprint are accumulated. Experimental results on challenging latent fingerprint datasets show the proposed method outperforms previous ones markedly.
Model-based vision is firmly established as a robust approach to recognizing and locating known rigid objects in the presence of noise, clutter, and occlusion. It is more problematic to apply model-based methods to images of objects whose appearance can vary, though a number of approaches based on the use of flexible templates have been proposed. The problem with existing methods is that they sacrifice model specificity in order to accommodate variability, thereby compromising robustness during image interpretation. We argue that a model should only be able to deform in ways characteristic of the class of objects it represents. We describe a method for building models by learning patterns of variability from a training set of correctly annotated images. These models can be used for image search in an iterative refinement algorithm analogous to that employed by Active Contour Models (Snakes). The key difference is that our Active Shape Models can only deform to fit the data in ways consistent with the training set. We show several practical examples where we have built such models and used them to locate partially occluded objects in noisy, cluttered images.
The theoretical framework and algorithms of a novel method for the generation of synthetic on-line signatures are presented. This model-based approach combines the spectral analysis of real signatures with the Kinematic Theory of rapid human movements in order to generate totally synthetic specimens. Two different algorithms are also described in order to produce duplicated samples from the synthetic master signatures, so that the generation scheme as a whole is able to produce in a complete automatic fashion huge synthetic databases. Typical examples of synthetic specimens are presented to highlight their human-like appearance. The validation protocol and the test results are presented and discussed in a companion paper.