Person Invariant Classification of Subtle Facial Expressions Using Coded Movement Direction of Keypoints

摘要

This paper describes a person invariant method for classifying subtle facial expressions. The method uses keypoints detected by using a face tracking tool called Face Tracker. It describes features such as coded movements of keypoints and uses them for classification. Its classification accuracy was evaluated using the facial images of unlearned people. The results showed the average F-measure was 0.93 for neutral (expressionless) facial images, 0.73 for subtle smile images, and 0.92 for exaggerated smile images. Also, person invariant accuracy was evaluated by using F-measure frequency of unlearned people. The results revealed that the proposed method has higher person invariant accuracy than the previous methods.