Effect of Facial Expression Categories and Calculation Methods on Automatic Emotion Recognition

摘要

Automatic classifiers are becoming increasingly popular for emotion recognition purposes. In order to compare the accuracy of seven commercially available classifiers, 468 video recordings of posed facial expressions supposedly displaying anger, disgust, fear, happiness, sadness or surprise were processed. Four different calculation methods to identify the emotion recognised were compared. A generalised linear mixed model reveals significant differences in automatic classifiers, in the type of emotions displayed, and in calculation methods. Some emotions such as happiness, sadness and surprise appear to lead to relatively high recognition accuracy, whereas anger, disgust and fear lead to low recognition accuracy. Finally, differences between the calculation methods are observed, as the Matching Score method indicates lower accuracy compared to the three other methods. These results support the necessity to establish standards for automatic classification of facial expression.