Characteristics of LBP-TOP Feature for Emotional Estimation by Micro-expression

Bad news, such as cancer notifications from doctors, has a big impact on patients. The patient does not remember the contents of the interview. Therefore, it is necessary for medical doctors to estimate the patient's emotions during the interview and to understand the patient's internal condition. In the medical field, the development of a communication support system is required. The purpose of this study is to improve the accuracy of emotional estimation using micro-expressions which would express true emotions. The emotional estimation employs the feature selected LBP-TOP (Local Binary Patterns from Three Orthogonal Planes) using the ratio of the inter-class variance to within class one. In this paper, we show the characteristics of emotional classification using LBP-TOP feature based on the experimental results obtained with the XGBoost and SVM (Support Vector Machine).