Motion magnification multi-feature relation network for facial microexpression recognition

Abstract Microexpressions cannot be observed easily due to their short duration and small-expression range. These properties pose considerable challenges for the recognition of microexpressions. Thus, video motion magnification techniques help us see small motions previously invisible naked eye. This study aimed enhance microexpression features with amplification technology. Also, a multi-feature relation network (MMFRN) combining two feature modules was proposed. The spatial is enlarged while completing extraction, which used classification. In addition, we transferred Resnet50 extract global improve comprehensiveness. controlled through hyperparameter factor α. effects different factors on results are compared, best selected. experiments have verified that can resolve misclassification problem caused by one-to-one correspondence between microexpressions facial action coding units. On CASME II datasets, MMFRN outperforms traditional methods other neural networks.