Automatic 3D facial expression recognition using geometric scattering representation

Facial Expression Recognition (FER) is one of the most active topics in the domain of computer vision and pattern recognition, and it has received increasing attention for its wide application potentials as well as attractive scientific challenges. In this paper, we present a novel method to automatic 3D FER based on geometric scattering representation. A set of maps of shape features in terms of multiple order differential quantities, i.e. the Normal Maps (NOM) and the Shape Index Maps (SIM), are first jointly adopted to comprehensively describe geometry attributes of the facial surface. The scattering operator is then introduced to further highlight expression related cues on these maps, thereby constructing geometric scattering representations of 3D faces for classification. The scattering descriptor not only encodes distinct local shape changes of various expressions as by several milestone descriptors, such as SIFT, HOG, etc., but also captures subtle information hidden in high frequencies, which is quite crucial to better distinguish expressions that are easily confused. We evaluate the proposed approach on the BU-3DFE database, and the performance is up to 84.8% and 82.7% with two commonly used protocols respectively which is superior to the state of the art ones.