TAG: A two-level framework for user authentication through hand gestures

Gesture-based interfaces are gaining attention from researchers worldwide owing to a user-friendly and natural mode of interaction. The direct use of hands as an input method is an attractive way to design Human-Computer interfaces. In scenarios where hand gestures are used to perform manipulation tasks on sensitive data, there is a need to authenticate a user. In this paper, we propose a two-level framework for addressing this need by augmenting the existing gesture recognition methodologies with hand biometrics. In the first level, the hand gesture is recognized by extracting Gesture-Specific Features (GSFs) from a video consisting of a gesture password. This is followed by the second level where the user is authenticated based on User-Specific Features (USFs). These GSFs and USFs together form a Feature Vector (FV) which is unique for every user. This is fed into LIBSVM for authentication purposes. Before an unknown or a malicious user tries to modify sensitive data, his presence is detected, he is denied access and a warning is sent to the registered user. The framework is tested using hand gestures derived from American Sign Language symbols. An overall accuracy rate of 95 percentage demonstrates the potentiality of TAG in access control through a hand gesture interface.