562015-03-24 , 法政大学大学院理工学・工学研究科
My research goal is to achieve accurate and natural automatic facial expression recognition by examining features that reveal the dynamic deformation of three-dimensional (3D) faces. We conducted classification experiments to test whether the geometric features obtained by a motion capture system could be used to discriminate among different categories of facial expressions. Classification experiments use the simple minimum distance classification method based on Euclidean distance, and Leave-one-out method is used to the method of evaluation. We discovered that geometric features representing 3D coordinates of sparse feature points defined on a subject’s face could be measured by a motion capture system for each time frame. The measurements of these features are mostly effective at discriminating different facial expressions, especially at time frames that reveal maximum expressiveness. However, the generation of facial expressions includes a series of temporal transitions of feature points from an initial emotionless face to an expression’s peak. Therefore, this study examines the effect of the temporal integration of gradual 3D deformation into features used for facial expression discrimination.