The classification of facial expressions with multi-cultural backgrounds: an event-related fMRI study

Expression recognition is important for our social interaction and communications, but the role of face-selective regions in discriminating various facial expressions remain unclear, especially when the expressions came from multi-cultural backgrounds. In this study, 800 facial expressions collected from 5 different facial expression databases with western or eastern cultural backgrounds were shown to the subjects in a slow event-related fMRI experiment. The subjects were instructed to indicate the category of facial expressions (happy, disgust, angry or neutral) by pressing different buttons. One multivariate pattern analysis method, support vector machine was trained to predict the categories of facial expressions. Results showed that: (1) the face selective regions differed in their ability for expression decoding, but a similar pattern was observed, with a predominance to classify facial expressions with opposite valence, i.e. happy vs. fear and happy vs. disgust. Besides,angry vs. disgust and happy vs. neutral achieved the lowest results. (2) the accuracies of facial expression classification cross-databases were as high as the accuracy of the generalization across runs within-database. These results provided evidence for the consistency of the representation of facial expressions in human brain with different culture backgrounds.