3D facial expression recognition using delta faces

Facial expression recognition has attracted a lot of attention due to its potential applications in human-computer interaction as well as in human facial behavior research. In the past decades, its research is based on 2D images or videos, and it has been recently explored into 3D spaces to solve the problems caused by illumination and pose variation. Since 2006, 3D facial expression recognition (3D FER) has been greatly fostered with the creation of BU-3DFE (Binghamton University 3D Facial Expression) database. However, as this database provides not only 3D facial models but also a dense set of 83 manually labeled landmarks for each model which leads to many of previous algorithms for 3D FER are semi-automatic. This paper introduces delta faces to set 3D FER fully automatic as the delta faces are achieved directly from 3D mesh models and a reference neutral expression. Seen from experimental results, especially the result of LBP+SVM (91.3%), which is far beyond previously published methods, this strategy for 3D FER are verified to be remarkable.