Signal Processing - July 2016 - 58

Table 5. Key multimodal fusion-based continuous authentication methods for mobile devices.
Study
Zhang et al. [61]
Fridman et al. [62]
Shi et al. [63]
McCool et al. [65]
Sitova et al. [66]
Saevanee et al. [67]

Modalities
Face, touch gestures
Stylometry, application usage, web
browsing, GPS location
Accelerometer, touch screen, microphone, location history
Face, voice
HMOG, tap, keystroke

Number of Users
50
200

Fusion Method
Feature level
Decision level

Performance (%)
RR: 83.75
EER: 5 (One minute), 1 (30 minute)

Seven

Decision level

EER: -

152
100

Score level
Score level

HTER: 11.9 (male), 13.3 (female)
EER: 7.16 (walking), 10.05 (sitting)

Linguistic profiling, behavioral profiling,
keystroke dynamics

Score level

EER: 3.3 (weighting) - 4.4 (sum)

the security level of the continuous authentication system,
within a continuous authentication framework. Several
while a higher false rejection rate will frustrate a legitimate
research studies have specifically focused on the applicabiluser, which is less dangerous than a lower security level. It
ity of these biometrics modalities for nonintrusive authenwas argued in [70] that to be able to build reliable, effectication. It is seen that physiological biometrics such as the
tive, and usable systems, one needs specific guidelines that
face can provide higher authentication accuracy than behavtake into account the specific constraints of security mechaioral biometrics such as gait or touch gestures. Further, as
nisms. Furthermore, security systems should be built so
behavioral biometric characteristics tend to change over
as to be easy to learn and use by users with different backtime and under various environmental conditions, one has to
grounds and skills. It was also argued that human factors
constantly update the templates to maintain the performance
should be incorporated into the design of
of these techniques. The tradeoff among
continuous authentication systems, where
computation, processing speed, and accuusability is central during the whole develracy has to be considered when using
Physiological biometrics
opment process.
these modalities for transparent authentisuch as the face
Several works have discussed the issue
cation. For example, the face-based concan provide higher
of usability of continuous authenticatinuous authentication system requires
tion systems. For instance, in [68] a proone to detect, align, and recognize faces
authentication accuracy
totype was developed using keystroke,
from the images or videos collected from
than behavioral
voice, and face biometrics for continuous
the front-facing camera. Each of these
biometrics such as
authentication. The prototype was evalusubalgorithms can be very time consumgait or touch gestures.
ated using 27 participants, and the study
ing, making the overall matching algoreported that 92% of the participants
rithm computationally demanding and not
considered it more secure in comparison
real time. In contrast, touch gesture-based
to the traditional methods of authentication. Similarly, [69]
methods often do not require detection or segmentation of
conducted an in-lab study of security perception of implicit
data. Hence, they could be more efficient in terms of proauthentication with 30 users based on behavioral biometcessing speed. It can be concluded that there is not a single
rics. In their study, the researchers asked users to complete a
biometric modality that is ideally suited for all scenarios.
series of tasks on a smartphone that was ostensibly protected
However, a significant amount of prior research has shown
with varying degrees of transparent authentication. They
that continuous authentication methods based on multiple
then surveyed the participants regarding their opinion about
biometric traits are often superior to unimodal continuous
transparent authentication. They found that 73% of particiauthentication systems.
pants felt that implicit authentication based on behavioral
biometrics was more secure than traditional methods such
Usability and security issues
as PINs and passwords and 90% indicated that they would
The usability of transparent continuous authentication sysconsider using a transparent authentication method on their
tems on mobile devices has become a major issue in research
own mobile device.
[5], [68], [69]. A balance needs to be struck between security
More recently, a two-part study consisting of a controlled
and usability of a biometrics-based continuous authenticalab experiment and a field study was conducted in [5] on
tion system. The design of usable yet secure continuous user
implicit authentication usability and security perceptions with
authentication systems raises crucial questions concerning
37 participants. The study indicated that 91% of participants
how to solve conflicts between mobile security and usabilfound implicit authentication to be convenient and 81% perity. For instance, in the continuous authentication context, a
ceived the provided level of protection to be satisfactory. Furfalse rejection is less costly than a false acceptance. This is
thermore, the authors found that false accepts and detection
due to the fact that higher false acceptance rates will lower
58

IEEE Signal Processing Magazine

July 2016

Table of Contents for the Digital Edition of Signal Processing - July 2016