3D Nasal Shape: A New Basis for Soft-Biometrics Recognition

In the past 10 years, Soft-Biometrics recognition using 3D face has become prevailing, with many successful research works developed. In contrast, the usage of facial parts for Soft-Biometrics recognition remains less investigated. In particular, the nasal shape contains rich information for demographic perception. They are usually free from hair/glasses occlusions, and stay robust to facial expressions, which are challenging issues 3D face analysis. In this work, we propose the idea of 3D nasal Soft-Biometrics recognition. To this end, the simple 3D coordinates features are derived from the radial curves representation of the 3D nasal shape. With the 466 earliest scans of FRGCv2 dataset (mainly neutral), we achieved 91% gender (Male/Female) and 94% ethnicity (Asian/Non-asian) classification rates in 10-fold cross-validation. It demonstrates the richness of the nasal shape in presenting the two Soft-Biometrics, and the effectiveness of the proposed recognition scheme. The performances are further confirmed by more rigorous cross-dataset experiments, which also demonstrates the generalization ability of propose approach. When experimenting on the whole FRGCv2 dataset (40% are expressive), comparable recognition performances are achieved, which confirms the general knowledge that the nasal shape stays robust during facial expressions.