Comparative Study of Feature Extraction Method for Emotional Classification by Micro-expressions

Human facial expressions include a slight and instantaneous movement called micro-expressions. Unlike ordinary facial expressions, micro-expressions are impossible to control by oneself. Since micro-expression shows true emotions, micro-expression recognition is expected to play an active role in clinical diagnosis and business negotiations. However, it is difficult to recognize micro-expression because of their insensible and quick facial movements. In this study, we aimed to improve the accuracy of emotional estimation using micro-expressions. In previous researches on emotional estimation using micro-expression, LBPTOP (Local Binary Pattern from Three Orthogonal Planes) and CBP-TOP (Centralized Binary Patterns from Three Orthogonal Planes) have been utilized. However, it is unclear if the feature selection and the combination of multiple features for emotional classification are effective. In this study, the emotional classification was performed using selected components of each individual feature. In addition, we investigated whether the fusion of scores obtained from each feature improved the accuracy of emotional estimation. The experimental results showed that the accuracy of emotional classification was increased by feature selection, whereas the score level fusion did not contribute to improve the performance of emotional estimation.