Facial behaviometrics: The case of facial deformation in spontaneous smile/laughter

In this paper we explore the use of dense facial deformation in spontaneous smile/laughter as a biometric signature. The facial deformation is calculated between a neutral image (as neutral we define the least expressive image of the smile/laughter episode) and the apex of spontaneous smile/laughter (as apex we define the frame of the maximum facial change/deformation) and its complex representation is regarded. Subsequently, supervised and unsupervised complex dimensionality reduction techniques, namely the complex Principal Component Analysis (PCA) and the complex Linear Discriminant Analysis (LDA), are applied at the complex vector fields for feature extraction. We demonstrate the efficacy of facial deformation as a mean for person verification in a database of spontaneous smiles/laughters.