Automated Robust Facial Expression Recognition using Transfer Learning ResNet50

Abstract

The human face is a convenient, fast, and accurate source of communication.  Facial expressions convey internal human emotion developed using different facial traits. Affective computing works on developing systems for Facial Expression Recognition (FER) using machine learning tools and it remains an active research area for the research community.  This paper proposes a deep learning-based model ResNet50 for facial expression recognition. We further applied transfer learning and fine-tuning techniques with the proposed model to improve the generalization. The model is trained and validated at the FER2013 dataset and tested with some unseen images from MMA facial expression dataset. The model archives validation accuracy of 86.32% which is comparable with existing research.