Privacy Preserving Transparent Mobile Authentication
Julien Hatin
1,2
, Estelle Cherrier
2
, Jean-Jacques Schwartzmann
1,2
and Christophe Rosenberger
2
1
Orange Labs, 14000 Caen, France
2
Normandie Univ., ENSICAEN, UNICAEN, CNRS, GREYC, 14000 Caen, France
Keywords:
Transparent Authentication, Privacy, Cancellable Biometric, Smartphone, Online Authentication.
Abstract:
Transparent authentication on mobile phones suffers from privacy issues especially when biometric infor-
mation is involved. In this paper, we propose a solution to address those two issues using the Biohashing
algorithm on behavioral information extracted from a mobile phone. The authentication scenario is tested on
a dataset composed of 100 users and shows promising results with a 10% EER in the worst case scenario (i.e
when protection key is compromised) and a 1% EER in the best case one. In addition, privacy concerns are dis-
cussed and experimentally evaluated both in a quantitative and qualitative ways. This opens new perspectives
concerning online authentication using smartphone sensing abilities.
1 INTRODUCTION
Mobile phone security is becoming a predominant is-
sue as more and more online applications and services
are now available for smartphones. Often those appli-
cations consider having the smartphone is sufficient
to access the service (Grosse and Upadhyay, 2013)
and use the remember me function. When more se-
curity is needed, or when it is needed to authenti-
cate on another device than the smartphone, burden is
added to the user by asking him to remember a pass-
word. To minimize this burden, users often choose
the same password for multiple services and easy-to-
guess passwords like 123456 (Vance, 2010). An alter-
native is to use transparent authentication. It permits
to seamlessly authenticate an user. To explain what
transparent authentication is, we quote Nathan Clarke
(Clarke, 2011):
Definition 1 (Transparent Authentication). Transpar-
ent authentication can be achieved by any authenti-
cation approach that is able to obtain the sample re-
quired for verification non-intrusively.
Behavioral biometrics is a perfect candidate for
transparent authentication. Indeed, in this approach,
authentication is based on the way someone does
something. The authentication burden is removed
when using behavioral biometrics because samples
are recorded seamlessly.
In May 2016, Google announced the release of a
continuous and transparent authentication mechanism
to replace the {login, password} couple (Google,
2016). This solution is announced to be available by
the end of the year as an API. However, concerns still
occur (Patel et al., 2016) : (i) Often this kind of au-
thentication are outsourced to companies that have the
expertize in the field of transparent authentication and
identity management; (ii) Unlike passwords, biomet-
rics cannot be cancelled.
Besides, samples can be of any type including ge-
olocation, application usage, web browsing history,
emailing and many more, therefore important privacy
concerns occur. This is even more true when the au-
thentication task is outsourced. In addition, the can-
cellability of biometric data is an important security
risk in case of theft of a large dataset and is an im-
portant brake to the deployment of online transparent
authentication services.
Our contribution is a new authentication frame-
work that is privacy preserving and cancellable. In
addition, this new scheme does not add any burden
to the user and can by applied on multiple samples
issued from different smartphone sensors. The sys-
tem is evaluated on a dataset containing data from 100
users collected during one month.
The organisation of the paper is as follows. We
first describe the related works concerning behavioral
authentication on mobile devices in section 2. Sec-
tion 3 proposes a framework that uses behavioral sam-
ples and the Biohashing algorithm to protect privacy
and make the data cancellable. Section 5 rounds off
the paper with a conclusion and indications of future
works.
354
Hatin, J., Cherrier, E., Schwartzmann, J-J. and Rosenberger, C.
Privacy Preserving Transparent Mobile Authentication.
DOI: 10.5220/0006186803540361
In Proceedings of the 3rd International Conference on Information Systems Security and Privacy (ICISSP 2017), pages 354-361
ISBN: 978-989-758-209-7
Copyright
c
 2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

2 RELATED WORKS
This part is dedicated to a short states of the art both
for behavioral authentication and mobile authentica-
tion.
Behavioral authentication solutions that provide
transparent authentication are a fast growing area.
This is especially due to the Active Authentication
project (Guidorizzi, 2013). The Defense Advanced
Research Projects Agency offers to move beyond
password by using transparent authentication mecan-
ism. This means most users will authenticate them-
selves using biometrics sensors.
The authors of (Hayashi et al., 2013) proved that
combining the location with a standard authentication
increases the global trust in that authentication. In ad-
dition, this article shows that the main locations aris-
ing for a user are: (i) Home and (ii) Workplace. This
implies to continuously know where the user is and
therefore compromises users privacy. The location
property and especially the one offered by GPS sen-
sors embedded in modern smartphones represents dis-
criminating features. In reference (Das et al., 2013),
the authors record the daily activities (including loca-
tion) of the user to ask him questions about its past
day in order to authenticate himself. This implies to
store private data about the user.
The authors in (Li et al., 2013) offer a solution
to authenticate users using the geolocation and the
phone calls. They obtain an EER of 5.4% with the
6 last phone calls. However, the privacy aspect is not
taken into account. In (Saevanee et al., 2014), the
authors cumulate different authentication modalities
and also include the text message content. To proceed
with the text message information, the messages must
be read. This implies privacy leakage.
Less sensitive data can be exploited to perform
behavioral authentication. This is the case of gait
recognition (Derawi and Bours, 2013). However, the
authors in (Tanviruzzaman and Ahamed, 2014) have
shown that combining location information with gait
recognition increases the global performance of the
system. By combining those data, they obtained an
ERR of 10% on a dataset of 13 users. However, pri-
vacy protection is not taken into account. Another
approach is to use the swipe gesture patterns as pro-
posed in (Mondal and Bours, 2013). The touch dy-
namics is one of the most commonly used methods to
transparently authenticate users on smartphones.
Besides, mobile sensors are exploited to build
frameworks for user authentication purpose as in ref-
erence (Witte et al., 2013). A probabilistic approach
is performed in (Kayacik et al., 2014). In this paper,
the authors propose to store data on the phone. This
mitigates the privacy issues but the data saved on the
mobile phone become an issue when the device has
to be replaced: in this case, data have to be trans-
ferred or stored online. The authors in (Fridman et al.,
2015) propose a solution that uses multiple informa-
tions that can be extracted from the mobile phones
to monitor the behavior of the user. This monitoring
includes text messages, web browsing habits, applica-
tion usage and location of the user. The performances
are promising with a FAR around 11% and a FRR
around 6% for the location modality alone using an
SVM classifier. By merging this modality with other
behavioral modalities, authors are able to achieve an
EER of 5% after 1 minute of user interaction with
the device. Those results are obtained on a private
database of 100 users. No privacy protection scheme
is envisaged in this paper.
The privacy problem was noted in (Jakobsson
et al., 2009). In this article, the authors advise
to (i) remove unique identifier information, (ii) use
pseudonyms, (iii) use aggregated data.
To the best of our knowledge, there are few pro-
posed solutions in the literature to deal with privacy
concerns. The authors in (Safa et al., 2014) use an
homomorphic encryption scheme. In (Nauman et al.,
2013) the authors describe a protocol dedicated to
keystroke analysis. In (Chow et al., 2010), the authors
address the problem of online authentication using
implicit information and store the data directly on the
mobile phone, thus delegating the autorization server
role to the mobile phone. This permits to mitigate the
privacy problem but does not solve the cancellability
issue.
The main contribution of this paper is to propose
a solution for mobile authentication enabling privacy
protection and data cancellation.
3 PROPOSED APPROACH
The proposed solution merges different sensors infor-
mation from the mobile phone and use the Biohash-
ing algorithm to ensure both cancellability of sensi-
tive data and privacy protection.
3.1 Biohashing
Biometric data are personal data, intrinsically non-
revocable (unlike passwords, or tokens) and thus very
sensitive data. The BioHashing algorithm is one par-
ticular example of cancelable biometrics techniques
(Ratha et al., 2001; Bolle et al., 2002). The concept
of cancelable biometrics relies on a transformation of
the raw biometric data, enabling the transformed data
Privacy Preserving Transparent Mobile Authentication
355

to address both security and privacy protection issues.
The general principle consists in the generation of a
new biometric template, from the biometric feature
vector and a secret random number. Therefore, it can
be seen as a two-factor authentication scheme. Mo-
bile authentication is particularly suited to perform
cancelable biometrics based authentication: the mo-
bile can easily capture various biometric modalities
and at the same time the secret random number can
be stored in the secure element (see section 3.2.2). As
every biometric system, cancelable ones involve two
steps.
• Enrolment step: once transformed, the new tem-
plate (or transformed template) is stored as refer-
ence in the mobile phone, while the original raw
biometric vector is discarded and never kept.
• Verification step: a comparison is performed be-
tween two transformed templates, namely be-
tween the transformed query and the transformed
reference.
For an overview of the other existing cancelable bio-
metrics schemes, we refer the reader to many survey
papers (Rathgeb and Uhl, 2011), (Patel et al., 2015).
BioHashing has been first described in (Goh and Ngo,
2003) and (Teoh et al., 2004), it has been respectively
applied to face recognition and fingerprint recogni-
tion. The principle is as follows. The transforma-
tion function in BioHashing combines a user-specific
secret key K with the biometric feature expressed as
a fixed-length vector x = (x
1
, . . . , x
n
) ∈ R
n
. For more
protection, the key K is stored in the secure element of
the mobile phone (see section 3.2.2). The BioHashing
process is divided into two steps:
• Random projection: the key K is used as a
seed to generate m random vectors r
j
∈ R
n
, j =
1, . . . , m and m ≤ n. After orthonormalization
by the Gram-Schmidt method [37], these vec-
tors are gathered as the column of a matrix O =
(O
i, j
)
i, j∈[1,n]×[1,m]
. A projection of the biometric
feature vector ( f
1
, . . . , f
n
) is then computed.
• Quantization: This step is devoted to the transfor-
mation in a binary-valued vector of the previous
real-valued vector using a simple thresholding.
More precisely, a binary vector B = (B
1
, . . . , B
m
)
called BioCode is obtained from the previous pro-
jected vector by thresholding. The goal of this
step is to guarantee the irreversibility of the whole
process. It requires the specification of a threshold
to compute the final BioCode.
Thus, by combining the high confidence of the key to
the biometric data, the inter-class variation increases
while the intra-class distance is preserved. Hence,
good performances can be achieved, see (Belguechi
et al., 2013a) for example.
We insist on the fact that BioHashing is privacy-
preserving. Indeed, with BioHashing algorithm, as
well as other cancelable biometrics techniques, orig-
inal raw biometric data does not have to be stored.
Therefore the users privacy is guaranteed. Moreover,
revocability property is also automatically ensured:
if the transformed template is compromised, the se-
cret random key can simply be replaced to generate a
new BioCode. Concerning the remaining properties
among the aforementioned ones, unlinkability is also
guaranteed together with the diversity, since distinct
secret keys can be chosen for distinct applications,
with no link between the generated BioCodes. Again,
these properties have been deeply analyzed (for fin-
gerprints) in (Belguechi et al., 2013a) and (Belguechi
et al., 2013b).
In the context of the present paper, i.e. for trans-
parent online mobile authentication, details are given
in the following section 3.2 about the architecture, i.e,
how we propose to implement BioHashing.
3.2 Architecture
The global architecture is composed of a client server
application. In this paper, the mobile phone collects
data about the user behavior when this one is calling.
The proposed approach can be extended to other sam-
ples from different sensors from the one used in this
work. It particularly well fits for merging geolocation
samples with any other sensors or group of sensors.
3.2.1 Client Architecture
Smartphones have a great penetration into the market
(Sophos, ) and possesses heavy sensing capabilities.
Within those sensors, we can find: accelerometers,
gyroscopes, light sensors, GPS and many others. In
addition, features can be extracted using the software
capabilities of the phone such as the call duration or
the applications usage.
In this work, we focus on the association of the
location with call information. Location can be com-
puted from the networks informations, the wifi net-
works available and the GPS of the mobile phone.
Call information can be extracted from the software
part of the phone and from hardware part too. For in-
stance, on the Android system it is possible to extract:
• The duration of the call
• The position of the phone (accelerometer, gyro-
scope)
• The location of the user
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
356

Figure 1: Client architecture.
• The phone number of the callee
• Noise informations from the environment
For this article, we have the following information
available in an dataset described in section 4:
• The location of the cell from where the call as
been made
• The phone number of the callee
Those data were chosen because they already offer
interesting performances in the literature (Li et al.,
2013). The generated Biocode size is dependent of
the input vector (see section 3.1. As a consequence,
the longer the input vector is, the better it is to avoid
collision between Biocodes. In this paper we only
use geolocation and phone number data because of
the available dataset. In a real use case, additional
data can be used.
The verification process is done online. Before be-
ing sent to the server, the content of the data must first
be anonymized. This step is performed using the Bio-
hashing algorithm. The privacy protection depends
on the security of the secret key used to perform the
Biohashing. To ensure this security, the key is stored
inside the secure element of phone. It can either be an
online secure element or a physical one.
Depending on the computing capabilities of the
secure element, the Biohashing algorithm could be
computed directly inside it. This way, the secret key
is never revealed. This is described in figure 1
Once the Biohashing is performed, the data can
be sent online. To avoid an interception of the bio-
hashed samples by an attacker, the data must be sent
through a secure channel. This could be done using
an TLS(Dierks, 2015) connection.
3.2.2 Server Architecture
The server receives BioCodes continuously each time
an user calls. The first step is to store the BioCode
Figure 2: Server architecture.
in the database. This database can be centralized be-
cause each user have its one key. The stored data per-
mits to construct a template. When enough data are
enrolled for an user, the server switch to verification
mode. This step is described in figure 2.
When in verification mode, the server keeps re-
ceiving BioCode from the mobile phone. When re-
ceiving a BioCode, the server uses a classifier to de-
termine whether the received sample is genuine or
not.
In the next section, we describe the experiments
we lead to evaluate the proposed approach.
4 EXPERIMENTS
4.1 Dataset
The collected dataset is a private real dataset contain-
ing the communication behavior of 100 users during
one month extracted from the network information.
The available data are:
• The latitude and longitude of cell
• The phone number of the caller
• The phone number of the callee
• The type of communication (Text message or
phone call)
We use those properties to perform a static and trans-
parent authentication. The idea is to use the behav-
ioral data extracted from a phone communication to
authenticate a user. To perform this authentication,
we choose not to use impostors data. This constraint
reflects real life industrial datasets. Even if those
datasets can contain millions of customers, impostors
data are not necessarily inside the dataset.
Each entry of the dataset is composed of a call or
a text message associated to the position of the user.
Table 1 presents the general statistics of the dataset.
We split the one-month dataset into two sets: the first
set, corresponding to the 15 first days is dedicated to
Privacy Preserving Transparent Mobile Authentication
357

the creation of the template, while the remaining data
are used for test purpose.
Table 1: Size of the communication dataset.
Enrollment data Verification data
Maximum 666 666
Minimum 16 14
Mean 157.5 109
In the following section, we describe the experi-
mental protocol used to evaluate the proposed solu-
tion both in terms of performances and in terms of
privacy protection.
4.2 Protocol
In the proposed approach, we evaluate both the sys-
tem performance and the privacy protection.
To evaluate the performance of the system we use
the False Match Rate and the False Non Match Rate.
Two biometric samples of the same person are not
exactly the same. As a consequence, an error can
occur if a collected sample is too different from the
one stored. This error is called the False Non Match
Rate (FNMR) (Jain et al., 2004). Oppositely, two
persons could have a similar behaviour and as conse-
quence the collected samples could be so close that
the recognition system accepts the impostor as the
genuine user. This error is called the False Match Rate
(FMR) (Jain et al., 2004).
To evaluate the privacy protection of the proposed
solution we use a measure based on the Shannon en-
tropy. The theoretical entropy of a perfect BioCode is
equal to its length. When computing the real entropy
a dataset of BioCode, we observe that the results are
inferior to the theoretical entropy. The difference be-
tween these two measures is called the privacy leak-
age and express how much information is available
concerning the original data. This value is expressed
in bits.
Two classifiers are evaluated. The first classifier
used is a Classical Support Vector Machine (Boser
et al., 1992) answers to two class classification prob-
lems. In the studied case, we model each user’s be-
havior with a one-class SVM. As a consequence, we
use a One Class Support Vector Machine. The aim
of a One Class SVM is to decide if new samples are
within the previously enrolled ones. The implementa-
tion used the libSVM(Chang and Lin, 2011) for Mat-
lab. Additionally, we compute the sum of the dis-
tance to the distance of the nearest entries to the sam-
ples collected in the user template. The approach was
preferred over the Kmeans algorithm because of the
dataset, because the geolocation are already gathered
by cell.
In order to evaluate the performance of the dataset,
we first compute the verification without any privacy
protection. In this first case, the data are sent in clear.
Then, we use the best case scenario where impostors
try to perform a zero effort attack. In this attack, they
simply try to authenticate with their own BioCodes.
Finally, the worst case scenario is envisaged, in this
case, the phone is stolen and the attacker try to per-
form usual phone calls. Those results are exposed in
the following section.
4.3 Experimental Results
We evaluate our results both in terms of performance
and and of privacy protection for the differents sce-
narios.
4.3.1 Without Privacy Protection
The first testing scenario is to authenticate without us-
ing any privacy protection on the data. This permits to
evaluate the basic performance of classical algorithms
on our dataset.
We use the caller number as an identifier to verify
our result. The only difference in this scenario is that
the biohashing algorithm is not applied on the data.
We evaluate two classifiers to determine the best ap-
proach for the current data. The first one is a One
Class SVM with nu = 0.0001 and the second one is
based on the KNN classifier. Results are summarized
in tables 2 and 3.
Table 2: One Class SVM without privacy protection.
FRR (%) FAR (%)
29.54 1.23
Table 3: KNN without privacy protection.
Number of
neighbors
EER (%) Corresponding threshold
1 8.39 0.15
2 8.39 0.30
3 9.15 0.49
4 9.28 0.67
5 9.77 0.86
Results using the KNN classifier are close to those
we can found in the literature (Li et al., 2013). We ob-
serve a small difference that can be explained by the
fact only one class is used for classification. However,
the One Class SVM does not offer sufficient perfor-
mance to be used as an authentication system. Similar
results were observed using the One Class SVM with
the privacy protection. About the privacy constraints,
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
358

none is respected because a potential attacker can ac-
cess all the point of interest usable for this authentica-
tion system. This means all sensitive information are
leaked. In the next section, we evaluate how Biohash-
ing improves those results.
4.3.2 Best Case Scenario
The best case scenario represents the normal utiliza-
tion of the authentication protocol. In this scenario, a
zero effort attack is performed. The Biohashing algo-
rithm was applied as follows : the latitude and longi-
tude were wrote in digits using respectively a 8 digits
precision and a 6 digits precision, the phone number
was concatenated to obtain a 24 digits tuple. This tu-
ple is then converted to a 104 bits binary vector. This
vector is then Biohashed to obtain a 100 bits Biocode.
Table 4 and table 5 summarize the results obtained
using a OCSVM and KNN.
Table 4: One Class SVM in the best case scenario.
FRR (%) FAR (%)
35.19 0
Table 5: KNN in the best case scenario.
Number of
neighbors
EER (%) Corresponding threshold
1 1.04 0.30
2 1.10 0.62
3 1.09 0.95
4 1.16 1.29
5 1.19 1.63
The privacy protection is evaluated by looking at
how much bits of information are leaked when us-
ing this protection. The evaluation process measures
the difference between a perfect dataset of random
vectors and our results. We obtain an entropy of 96
bits. It means a privacy leakage of 4 bits. This repre-
sents a great advance in terms of privacy preservation
compared to the previously available solutions which
leaves the data in clear.
Results obtained with using the privacy protection
are greatly improved. With an EER as low as 1% and
the privacy protection it can be envisaged to use this
solution in an online dataset.
4.3.3 Worst Case Scenario
In the worst case scenario, the attacker knows the
key. The only possible solution is that the attacker
has stolen the mobile phone because the key is stored
inside a secure element. The performances are shown
in table 6 and table 7.
Table 6: One Class SVM in the worst case scenario.
FRR (%) FAR (%)
34.60 2.68
Table 7: KNN in the worst case scenario.
Number of
neighbors
EER (%) Corresponding threshold
1 10.45 0.23
2 10.16 0.47
3 10.65 0.72
4 10.69 0.98
5 10.76 1.24
Even in this scenario, the secret key is not acces-
sible to an attacker. This protects the data contained
in the dataset. Assuming an attacker possess both the
mobile phone and the dataset, an attacker can only
guess location to see if the Biocodes are similar to
obtain information about the location of an user. Of
course because the attacker possess the mobile phone,
he has access to the call history and the called num-
bers.
5 CONCLUSION AND FUTURE
WORKS
The ROC curves for the three evaluated scenarios are
exposed in figure 3. We observe a great improvement
when using the Biohashing algorithm in the nominal
case (best case scenario). A limitation to this study is
the number of available samples. But even with this
limitation due to the available datasets, it shows great
opportunity in privacy preserving mobile authentica-
tion. In the performance evaluation we focus on the
call and location information. Those information can
also be used combined with any other sensors and also
Figure 3: ROC curves in the different scenarios.
Privacy Preserving Transparent Mobile Authentication
359

Table 8: Comparison with other solutions.
Ref. No.
of
Users
Features Privacy Performance Revocable
Li et al.(Li et al.,
2013)
71 Location & call None EER:8.8% with 1 sam-
ple and EER:5.3% with 6
samples
7
Savaenee et
al.(Saevanee
et al., 2014)
30 linguistic analysis,
keystroke dynam-
ics and behavioral
profiling
None EER:3.3%. 7
Tanviruzzaman et
al.(Tanviruzzaman
and Ahamed,
2014)
13 gps location and gait None EER:10% 7
Fridman et
al.(Fridman et al.,
2015)
200 GPS None FAR:11% and FRR:6% 7
Fridman et
al.(Fridman et al.,
2015)
200 text message content,
gps location, applica-
tions, web browsing
None ERR:5% after 1 minute
and EER:1% after 30
minutes
7
Chow et al.(Chow
et al., 2010)
50 phone, browser, sms,
gps
Data are
stored only on
the phone
Legitimate user can per-
form around 90 actions
before being rejected by
the system while an im-
postor can only per-
formed 10 actions before
being locked out
7
Safa et al.(Safa
et al., 2014)
Not
pro-
vided
calls, location, Wi-Fi
access, visited web-
sites
3-round pro-
tocol between
the device and
carrier
Not provided 3
Proposed frame-
work
100 calls, location Biohashing :
4 bits privacy
leakage
EER:1.04% best case
scenario, EER:10.45%
worst case scenario
3
using the time (Kayacik et al., 2014). In this case, a
similar approach to both allow revocation and privacy
protection can be applied.
In a future work, we wish to combine this modal-
ity with others to add new modality to a Single Sign
On environement.
The proposed approach is compared to the liter-
ature in the table 8. In this table both the different
privacy protection mechanisms and the performances
of the system are compared.
REFERENCES
Belguechi, R., Cherrier, E., Rosenberger, C., and Ait-
Aoudia, S. (2013a). An integrated framework combin-
ing bio-hashed minutiae template and PKCS15 com-
pliant card for a better secure management of finger-
print cancelable templates. Computers & Security,
39:325–339.
Belguechi, R., Cherrier, E., Rosenberger, C., and Ait-
Aoudia, S. (2013b). Operational bio-hash to preserve
privacy of fingerprint minutiae templates. IET Biomet-
rics, 2(2):76–84.
Bolle, R., Connell, J., and Ratha, N. (2002). Biometric per-
ils and patches. Pattern Recognition, 35(12):2727–
2738.
Boser, B. E., Guyon, I. M., and Vapnik, V. N. (1992). A
training algorithm for optimal margin classifiers. In
Proceedings of the fifth annual workshop on Compu-
tational learning theory, pages 144–152. ACM.
Chang, C.-C. and Lin, C.-J. (2011). Libsvm: A library
for support vector machines. ACM Trans. Intell. Syst.
Technol., 2(3):27:1–27:27.
Chow, R., Jakobsson, M., Masuoka, R., Molina, J., Niu, Y.,
and Song, Z. (2010). Authentication in the clouds: A
framework and its application to mobile users. In Pro-
ICISSP 2017 - 3rd International Conference on Information Systems Security and Privacy
360

ceedings of the 2010 ACM workshop on Cloud com-
puting security workshop.
Clarke, N. (2011). Transparent User Authentication Bio-
metrics, RFID and Behavioural Profiling. Springer.
Das, S., Hayashi, E., and Hong, J. l. (2013). Exploring
capturable everyday memory for autobiographical au-
thentication. In Proceedings of the 2013 ACM inter-
national joint conference on UbiComp ’13.
Derawi, M. and Bours, P. (2013). Gait and activity recogni-
tion using commercial phones. Computers & Security.
Dierks, T. (2015). The Transport Layer Security (TLS) Pro-
tocol Version 1.2. RFC 5246.
Fridman, L., Weber, S., Greenstadt, R., and Kam, M.
(2015). Active authentication on mobile devices via
stylometry, application usage, web browsing, and gps
location. arXiv preprint arXiv:1503.08479.
Goh, A. and Ngo, D. (2003). Computation of cryptographic
keys from face biometrics. In Communications and
Multimedia Security, pages 1–13. LNCS 2828.
Google (2016). Google Abacus project.
http://www.androidcentral.com/project-abacus-
atap-project-aimed-killing-password. [Online;
accessed 10-July-2016].
Grosse, E. and Upadhyay, M. (2013). Authentication at
scale. Security & Privacy, IEEE, 11(1):15–22.
Guidorizzi, R. P. (2013). Security: Active authentication.
IT Professional, 15(4):4–7.
Hayashi, E., Das, S., Amini, S., Hong, J., and Oakley, I.
(2013). Casa: Context-aware scalable authentication.
In SOUPS ’13 Proceedings of the Ninth Symposium
on Usable Privacy and Security.
Jain, A. K., Ross, A., and Prabhakar, S. (2004). An intro-
duction to biometric recognition. Circuits and Systems
for Video Technology, IEEE Transactions on, 14(1):4–
20.
Jakobsson, M., Shi, E., Golle, P., and Chow, R. (2009).
Implicit authentication for mobile devices. In Hot-
Sec’09 Proceedings of the 4th USENIX conference on
Hot topics in security.
Kayacik, H. G., Just, M., Baillie, L., Aspinall, D., and Mi-
callef, N. (2014). Data driven authentication: On the
effectiveness of user behaviour modelling with mobile
device sensors. CoRR, abs/1410.7743.
Li, F., Clarke, N., Papadaki, M., and Dowland, P. (2013).
Active authentication for mobile devices utilising be-
haviour profiling. International Journal of Informa-
tion Security.
Mondal, S. and Bours, P. (2013). Continuous authentication
using mouse dynamics. In Biometrics Special Inter-
est Group (BIOSIG), 2013 International Conference
of the, pages 1–12. IEEE.
Nauman, M., Ali, T., and Rauf, A. (2013). Using trusted
computing for privacy preserving keystroke-based au-
thentication in smartphones. Telecommunication Sys-
tems, 52(4):2149–2161.
Patel, V. M., Chellappa, R., Chandra, D., and Barbello,
B. (2016). Continuous user authentication on mobile
devices: Recent progress and remaining challenges.
IEEE Signal Processing Magazine, 33(4):49–61.
Patel, V. M., Ratha, N. K., and Chellappa, R. (2015). Can-
celable biometrics: A review. IEEE Signal Processing
Magazine, 32(5):54–65.
Ratha, N., Connell, J., and Bolle, R. (2001). Enhancing se-
curity and privacy in biometrics-based authentication
system. IBM Systems J., 37(11):2245–2255.
Rathgeb, C. and Uhl, A. (2011). A survey on biometric
cryptosystems and cancelable biometrics. EURASIP
J. on Information Security, 3.
Saevanee, H., Clarke, N., Furnell, S., and Biscione, V.
(2014). Text-based active authentication for mobile
devices. In ICT Systems Security and Privacy Protec-
tion, pages 99–112. Springer.
Safa, N. A., Safavi-Naini, R., and Shahandashti, S. F.
(2014). Privacy-preserving implicit authentication. In
IFIP International Information Security Conference,
pages 471–484. Springer.
Sophos. Mobile usage. https://www.sophos.com/en-
us/press-office/press-releases/2013/03/mobile-
security-survey.aspx. [Online; accessed 10-July-
2016].
Tanviruzzaman, M. and Ahamed, S. I. (2014). Your phone
knows you: Almost transparent authentication for
smartphones. In Computer Software and Applica-
tions Conference (COMPSAC), 2014 IEEE 38th An-
nual, pages 374–383. IEEE.
Teoh, A., Ngo, D., and Goh, A. (2004). Biohashing: two
factor authentication featuring fingerprint data and to-
kenised random number. Pattern recognition, 40.
Vance, A. (2010). If your password is 123456, just make it
hackme.
Witte, H., Rathgeb, C., and Busch, C. (2013). Context-
aware mobile biometric authentication based on sup-
port vector machines. In Emerging Security Technolo-
gies (EST), 2013 Fourth International Conference on,
pages 29–32. IEEE.
Privacy Preserving Transparent Mobile Authentication
361