Skip to Main Content
This paper presents a novel probabilistic approach for augmenting the performance of a 3D face recognition system with information from continuous facial soft biometric traits. In particular, by estimating the distribution of the noise induced during the measurement of one or more soft biometric traits, the recognition score for a genuine user can be efficiently modelled as a conditional matching probability that takes into account both geometric and soft biometrics. Herein, the geometric traits are provided via a state-of-art 3D face recognition system, while the soft biometrics regard the distances between three facial nodal points (i.e. the eyes, the nose and the mouth). Experimental validation on a proprietary dataset of 54 subjects illustrates significant advances in both identification and authentication rates of the proposed method when compared to the 3D face recognition system.