Cross-Cultural Nuances of Micro-Expressions and Action Units: A Comparative Study

This study explores the interplay between cultural norms, expressive traits, and computational modeling in the analysis of micro-expressions (ME) and facial action units (AUs) across Asian and Caucasian groups. We first investigate the differences in expressive occurrences, including their frequency, intensity, and duration of the two ethnic groups, and then evaluate the representative computational models on their vulnerability of being impacted by the cross-cultural domain shift.One of our hypotheses is that AU detection models are less affected by the cross-cultural domain shift than the ME recognition models, as the AUs are more rigorously defined according to anatomics by the Facial Action Coding System (FACS). Experimental results on 4DME, SMIC, and BP4D datasets are supportive of the hypothesis. Future research should work towards building ethnically balanced datasets and developing culturally adaptive models, especially for ME recognition.