A Novel Seven-Class Facial Expression Recognition Method With Face Mask
Abstract: Since COVID-19, face masks have become essential in people’s lives. When people are wearing masks, it is incredibly difficult for traditional face recognition techniques to identify facial expression. This paper proposes a seven-class facial expression recognition dataset FM-FER2013 (Face-Masked FER2013). The FM-FER2013 dataset is based on the FER2013 dataset. I also proposed a facial expression recognition method with face mask, based on YOLOv5 for face detection and ResNet for facial expression recognition. The accuracy are about 60% on the seven-class FM-FER2013 dataset and 73% on the five-class FM-FER2013 dataset. And then, real-world test is performed, with accuracy around 65%. The proposed method’s effectiveness and robustness demonstrate its high value and provide a hint for real-world application. The code implement is shown in https://github.com/RoyMikeJiang/FM-FER2013.
Loading