LAUSR

Different methods of user identification and authentication are widely used in security systems. Touch screen-based methods are very convenient for users and do not require specialized equipment, although commonly used pattern locks alone provide a low level of security. Analysis of biometric features of performed gestures can greatly improve the security of touch screen-based identification and authentication, however typically a large amount of training samples is needed to build an accurate model. In this study a novel method based on Elastic Shape Analysis and covariance shrinkage is introduced to overcome this limitation. Two dataset (one with finger gestures and one with stylus gestures) were used. It is shown that 4 training samples are sufficient for identification among 24 people with accuracy of 84.7 percent using elastic Linear Discriminant Analysis. In user authentication, the area under the curve (AUC) score of 0.986 was reached using an elastic variant of Principal Component Analysis and 8 training samples per user. The approach can be easily adapted as a part of a touch-screen based user authentication or identification system.