introduction.tex

\section{Introduction}
    Smartphone has been prevalently used in our daily life for storing and transmitting private data, conducting online payment. By 2016, the number of smartphone users is forecast to reach 2.1 billion and is expected to pass 5 billion by 2019~\cite{worldwide2016smartphonUsage}. However, the security for mobile access control has became a non-negligible issue because of its ubiquitous nature. It is reported by loolout.com that smartphone is easily lost or stolen by an attacker as its small size, and there are nearly \$2.5 billion worth of devices were lost or stolen in 2011~\cite{lookout-survey}. This may result in user's privacy leakage and finance lost.

    Recently, human identification system based on biometrics has emerged as an usable and secure authentication approach for access control and identity recognition. Unlike traditional authentication methods like passwords or PINs, Biometric-based approach exploits individual's physiological and/or behavioral modalities to recognize user's identity. Moreover, passwords or PINs are hard to remember~\cite{DeAngeli:2005:PRW:1090412.1090419} and can be stolen by shoulder surfing~\cite{Kwon2014Covert,shoulder} or video-based attacks~\cite{shukla2014beware,yue2014blind} while individual's biometric traits are not easily to be stolen or forged.

    In general, biometric-based technologies can be categorized as two types: physiological characteristics and behavioral traits. Physiological characteristics based on the personal trait to verify a user. Behavioral traits based on the way people do things. \emph{Physiological biometrics} such as fingerprints~\cite{Chen2005A}, voice~\cite{Rose2007Method} and iris~\cite{Brunelli1995Person} have already been widely commercial used with a high identification rate. However, these bio-features have facing the potential risk of being replayed by attackers. For example, fingerprint can be easily replicated by fingerprint membrane and voice can be forged by professional voice processing software. Even some researchers stated that iris can be counterfeited using victim's picture acquired from social media. Moreover, physiological biometrics is non-revocable, which means biometrics would permanent leakage once it was stolen. \emph{Behavioral biometrics} have been explored by researchers in the past few years. Touch-based behavioral traits such as multitouch gestures~\cite{Sae2014Multitouch} and keystroke pattern~\cite{Zahid2009Keystroke} have been proposed to provide access control on smartphone. However, these methods are not appreciated by user as they require close interaction with the touch-screen. This is not convenient in a frequent use. Other behavior-based methods such as in-air signature~\cite{Bailador2011Analysis} and gait recognition~\cite{Wang2016Gait} have the potential risks for adversaries to mimic. In addition, gait pattern is easily be altered by both the road condition and people's mental state.

    Human facial expression plays an significant role in our social interaction. It is driven (motivated) by the complex interaction between the emotional state and the facial muscles~\cite{Fridlund1994Human}. Due to the fact that it carries both psychological and behavioral information, facial expression is unique so that it is highly immune to the replay attack by the attackers. Moreover, unlike fingerprint and iris, facial expression is revocable as an individual has many facial expressions such as six basic expressions (anger, disgust, fear, joy, sadness and surprise) and compound expressions constructed by combining basic expressions~\cite{Du2014Compound}. Thus, facial expression can be regarded as facial behavioral biometric modalities for authenticating identity on smartphone.

    In this paper, we present a novel facial behavior authentication system for smartphone based on the serval seconds facial expression footage. It analyzes the dynamic changes of the facial expression and extracts an unique facial behavioral modalities for recognizing user's valid identity. Firstly, the dynamic changes of facial expression is related to facial bio-structure and each individual has a unique facial muscle condition. Specific facial structure or muscle-related bio-features such as the facial deformation or the distance of facial features are highly individual-dependent. Secondly, facial expression usually highly complies specific mental activity of human beings. Studies have found the close relationship between facial expression and human emotions~\cite{Matsumoto2013Reading}. Therefore, individuals hold unique facial behavioral modalities during expressing their emotions due to their different habits and experiences.

    In our work, we develop an biometric-based authentication system and implement a prototype on android smartphone using video footage that captures the user's facial expression when authenticating the identity. Unlike \emph{EyeVeri}~\cite{Song2016EyeVeri}, our approach does not require the visual stimuli to extract the physiological and behavioral biometrics. The user can appear any expression he wants to do when unlocking the device. This is more convenient and usability than the \emph{EyeVeri} as the design of visual stimuli are obtrusive and require explicit action from the user. Furthermore, the video is filmed through built-in front camera of smatrphone. This differ from~\cite{Sluganovic2016Using} that needs a additional eye tracking device, which limits it from being applied to a mobile environment.

    This authentication system employs a computer vision algorithm to track the facial behavior from the video. Using the deformation information of facial features extracted from the facial behavior, it then establishes a disaggregated model to verify user's identity. Simultaneously, during authentication process we construct another classifier using the Gabor features of facial expression extracted by Gabor filter~\cite{Fogel1989Gabor} to improve the recognition accuracy.

    We thoroughly evaluate our approach using....

    \noindent \textbf{Contributions} The key contribution of this paper is a novel authentication system for smart devices. Our system exploits techniques developed in the computer vision domain to address the key challenges highlighted above.

    This paper makes the following specific contributions:

    \begin{itemize}
        \item \emph{A New System:} We propose and implement a security and usable biometric-based authentication system based on facial expression on the smartphone without any extra hardware. Furthermore, this system does not require any visual stimuli so that it can accomplish authentication within 2 or 4 seconds.
        \item \emph{New Findings: } We use the deformation information of facial features to authenticate user's identity and discovere that the facial deformation data can be regarded as a biometrics to uniquely identify one's identity.
        \item \emph{Exploring New Technologies:} We develop a new authentication method by combining the facial deformation data with the Gabor features to dual coordination for improving the recognition accuracy. Our comprehensive evaluation shows that we can recognize one's identity with a accuracy of above 90\%.
    \end{itemize}