Demystifying Facial Expression Recognition Using Residual Networks

Residual Networks (ResNets) have emerged as a significant breakthrough in computer vision, pattern recognition, and image processing, offering a new avenue for linking theories to practical applications. The development of ResNets Networks has greatly benefited the field of facial expression recognition (FER). ResNet incorporates the concept of skip connections and has become one of the most widely used neural networks. The paper provides a comprehensive overview of utilizing ResNet architectures to address the challenge of facial expression and emotion recognition. The authors specifically highlight several ResNet variants, including the 3D version of the Inception-ResNet layer combination, ResNet and atrous convolutions, dynamic geometrical image network with ResNet, facial expression recognition using conditional random fields and ResNet, FER using ResNet and heart rate variability observations, and ResNet with squeeze and exception networks.