中国科学院心理研究所机构知识库

Emotions can be recognized through facial expressions, but this becomes challenging when facial expressions are deliberately manipulated. This study introduces tissue oxygen saturation (StO2), a non-contact measured physiological signal of hemoglobin binding to oxygen, as a novel and expression-independent indicator for emotion recognition. Herein, the first StO2-emotion dataset is developed, which encompasses six basic emotions collected from 145 participants via emotional stimulus videos. Pattern analyses indicate that StO2 exhibits distinctive patterns under various emotions, whereas facial expressions have no impact on StO2. Based on this finding, we design the intensity difference pattern-masked feature (IMF) to amplify the changes in StO2 and construct the region of interest embedded Transformer network (ROI-EmbedFormer) to capture the synergistic changes across different facial regions. Finally, an accuracy of 80.31% is achieved for StO2-based emotion recognition. This work shows that StO2 is a promising signal that can identify genuine emotions even without facial expressions.