Effectiveness of Feature-selected LBP-TOP for Emotional Estimation by Micro-expression

Bad news given by a medical doctor induces a blank-headed situation and the patient does not remember the contents of an interview. Therefore, the importance of communication between the medical doctor and the patient is emphasized. For smooth communication, the medical doctor should understand emotions of the patient. In this study, we proposed an emotional estimation method using micro-expressions that cannot be controlled by oneself. In the proposed method, LBP-TOP (Local Binary Patterns from Three Orthogonal Planes) was adopted as the feature for emotional estimation. From the experimental results, the emotional classification rate was improved by introducing the feature selection with the ratio of the inter-class variance to within-class one. In addition, the ROI (Region of Interest) is defined as regions of a partial face for feature extraction. The features extracted from the ROI contributed to increasing the emotional classification rate compared to the feature extracted from the whole face.