Biometric Authentication Based on Face and Gesture Recognition: 'Wink twice if it's you?'

Abstract

It is possible to fool an authentication system based on face recognition with a picture of a person's face. This can be made impossible by introducing gesture recognition to the authentication process. In this thesis an authentication system is investigated that combines face and gesture recognition by instructing a user to perform a sequence of randomly determined gestures. In this system, every image in a stream of camera images is inspected to see if it contains a face. This is done by using the Viola-Jones object detection algorithm, which also looks for other regions of interest (the eyes and mouth in a face) if a face is found. After that, face and gesture recognition takes place based on the results of the Viola-Jones cascades along with the principal component analysis algorithm. If the recognised gestures correspond with the given instructions and face recognition is successful, authentication will succeed. Experiments showed that the Viola-Jones cascades performed their task well. In contrast, the performance of the gesture recognition process was poor and unreliable, as its performance was dependent on the subject using the system and different lighting conditions. This makes it difficult to use the proposed system and additional work is required to make the system suitable for real world usage.