Aspects of facial biometrics for verification of personal identity

This thesis studies various aspects of facial biometrics for the verification of personal identity in a multimodal framework. The research focuses on the mouth area and, more specifically, on the design of a lip tracking system for the extraction of visual features. The tracker is based on statistical chromaticity models and uses a B-spline representation of the contour of the lips. Shape variability is restricted to affine deformations of a linear combination of modes of shape variation, which are automatically estimated in a robust way using the tracking results provided by a first, rather unconstrained lip tracker. Tracking experiments were performed in a large multimedia database and the results were fed as input features to a Dynamic Time Warping algorithm for speaker verification purposes according to a published evaluation protocol. A weighted linear classifier is eventually trained for performing fusion experiments on the same database combining various verification modalities such as face and voice.