Eye Blink Based Detection
of Liveness in Biometric Authentication Systems
Using Conditional Random Fields
Mariusz Szwoch and Pawel Pieniażek
Gdansk University of Technology,
Department of Intelligent Interactive Systems, Poland
[email protected],
[email protected] Abstract. The goal of this paper was to verify whether the conditional
random fields are suitable and enough efficient for eye blink detection in
user authentication systems based on face recognition with a standard
web camera. To evaluate this approach several experiments were carried
on using a specially developed test application and video database.
Keywords: biometrics, liveness detection, anti-spoofing, conditional ran-
dom fields.
1 Introduction
Biometrics is an emerging technology that is used for human identification basing
on a set of physiological and behavioral human characteristics, such as finger-
prints, DNA, iris, face, voice timbre, or a style of writing [1]. As each of these
biometric features can be used alone, or in combination with others, to unam-
biguously identify a human, many different biometric systems have already been
developed for access granting to different resources such as bank accounts, re-
stricted areas, computer systems and others. Using biometric systems generally
easies and speeds up that access and also gains it up to another level where
access granting is not connected with passwords, tokens, or identity documents
but directly with a granted person.
Though, using biometrics in security and protection systems offers many ad-
vantages it can also have some drawbacks and threatens [2]. In order to be useful,
biometric systems must not only offer high reliability and positive recognition
rate but they should also be resistant to attacks of unauthorized persons. This
can be achieved in several ways such as using specialized hardware, forcing the
user to cooperate with a verification system, combining several multimodal at-
titudes, or creating more sophisticated and robust algorithms.
Rapid development of a computational power of digital systems allows for
creation of on-line biometric authentication systems operating on video stream
This work was supported in part by Polish National Science Centre research project
no. N N516 367936 and by departmental grant no. 020206.
L. Bolc et al. (Eds.): ICCVG 2012, LNCS 7594, pp. 669–676, 2012.
c Springer-Verlag Berlin Heidelberg 2012
670 M. Szwoch and P. Pieniażek
from a camera. Such systems uses face recognition methods as well as other
human facial features to locate and identify the observed person. One of the most
probable kind of attacks against such systems are spoofing attacks that use some
kind of hardware copy of human image such as photography, video recording, 3D
head models or even masks [2]. As the latter attitudes are somewhat troublesome,
getting a facial photography of a given person is quite easy, capturing by hidden
or telephoto camera or simply downloading from the Internet.
This paper presents an attempt to provide a reliable liveness detection sub-
system that could assist face recognition systems to resist spoofing attacks with
photographs. Its evaluation should point out whether such a subsystem can be
used in user authentication systems based on face recognition. The described at-
titude does not require any additional hardware, except of standard web camera,
nor any collaboration of a user. Instead, the algorithm uses some preprocessing
and segmentation results from the face recognition process in order to improve its
efficiency and reliability. The proposed attitude is presented in the next chapter.
Detection of Face Liveness. Liveness detection may be a significant factor
in user authentication systems for access granting [3]. In normal situation this
task is very easy for a human observing somebody’s face with depth of view
(DOV) information, possibility of combining other information such as voice and
comprehensive knowledge about human’s physiological face activities. However,
liveness recognition becomes harder for computer systems when scene is observed
with only one camera with narrowed field of view and no DOV information.
There have been proposed several attitudes to create anti-spoofing subsystem.
Some of them use additional or specialized hardware such as high-resolution
camera [4] to detect smooth appearance changes of a real face, additional near
infrared camera to detect facial vein map or thermograph infrared camera to
detect correct thermal map of the face [5]. These attitudes give good and reliable
results in most cases but cannot be applied on standard platforms using ordinary
web cameras. Another possibility is combining video and audio input into multi-
modal subsystem [6] or asking a user to do some facial activities, such as turning
his or her head or closing eyes. Unfortunately this forcing of collaboration may
be not convenient for the user.
Finally, in the third approach, some specific characteristics of live person in
video recording are investigated to distinguish real face from a photo. Some at-
tempts focus on recreation of the third dimension of a scene by analysis of the
head movements or the face illumination. Unfortunately, these attitudes are sen-
sitive to lighting conditions. Other attitudes concern temporal human activities
such as natural head moving and turning, facial expressions, and blinking. In
this paper just the eye-blinking human activity is used to confirm liveness in
video image input that was proposed in [7].
Eye-Blink Characteristics. Eye-blink is a physiological activity of closing and
opening the eyelids. Though the blink frequency may vary in quite wide range
depending on many elements the time of spontaneous blinking lasts about 250ms
[7]. As the web camera captures video at the minimum frequency of 15 fps, and
