Features extraction scheme for behavioral biometric authentication in touchscreen mobile devices

Common authentication mechanisms in mobile devices such as passwords and Personal Identification Number have failed to keep up with the rapid pace of challenges associated with the use of ubiquitous devices over the Internet, since they can easily be lost or stolen. Thus, it is important to develop authentication mechanisms that can be adapted to such an environment. Biometric-based person recognition is a good alternative to overcome the difficulties of password and token approaches, since biometrics cannot be easily stolen or forgotten. An important characteristic of biometric authentication is that there is an explicit connection with the user's identity, since biometrics rely entirely on behavioral and physiological characteristics of human being. There are a variety of biometric authentication options that have emerged so far, all of which can be used on a mobile phone. These options include but are not limited to, face recognition via camera, fingerprint, voice recognition, keystroke and gesture recognition via touch screen. Touch gesture behavioural biometrics are commonly used as an alternative solution to existing traditional biometric mechanism. However, current touch gesture authentication schemes are fraught with authentication accuracy problems. In fact, the extracted features used in some researches on touch gesture schemes are limited to speed, time, position, finger size and finger pressure. However, extracting a few touch features from individual touches is not enough to accurately distinguish various users. In this research, behavioural features are extracted from recorded touch screen data and a discriminative classifier is trained on these extracted features for authentication. While the user performs the gesture, the touch screen sensor is leveraged on and twelve of the user‘s finger touch features are extracted. Eighty four different users participated in this research work, each user drew six gesture with a total of 504 instances. The extracted touch gesture features are normalised by scaling the values so that they fall within a small specified range. Thereafter, five different Feature Selection Algorithm were used to choose the most significant features subset. Six different machine learning classifiers were used to classify each instance in the data set into one of the predefined set of classes. Results from experiments conducted in the proposed touch gesture behavioral biometrics scheme achieved an average False Reject Rate (FRR) of 7.84%, average False Accept Rate (FAR) of 1%, average Equal Error Rate (EER) of 4.02% and authentication accuracy of 91.67%,. The comparative results showed that the proposed scheme outperforms other existing touch gesture authentication schemes in terms of FAR, EER and authentication accuracy by 1.67%, 6.74% and 4.65% respectively. The results of this research affirm that user authentication through gestures is promising, highly viable and can be used for mobile devices