Normalisation and Recognition of 3D Face Data Using Robust Hausdorff Metric

In this paper a novel approach to recognising normalised 3D face data is proposed. A robust Hausdorff metric is applied to 3D images from the Face Recognition Grand Challenge v1.0 database. The proposed variant of the Hausdorff distance metric is completely data driven and enforces a consistency check on how well the data between any two faces correlate. This recognition algorithm is combined with a normalisation procedure that retains the integrity of the original 3D data and only uses the statistics of the available 3D face surfaces to identify invalid points and perform registrations. Experiments show that the robust Hausdorff metric combined with an effective normalisation procedure results in 3D face surfaces which are significantly more discriminable. Tests performed on the FRGC database produced a rank one recognition rate of 98.96% and an equal error rate of 1.6%.