A Survey on Deep Facial Expression Recognition

Abstract

This paper reviews state-of-the-art developments in facial emotion recognition using deep learning. We also have analyzed the performance and limitations of the reviewed state-of-the-art deep learning architectures and algorithms for facial emotion recognition. We discuss the contribution, model performance, and limitations of various architectures like CNNs and RNNs. We have also reviewed the performance of various state-of-the-art deep learning algorithms for facial emotion recognition. We discuss the contribution, model performance, and limitations of each algorithm. We have found that the most successful architectures are Hybrid CNN-RNNs, which combine convolutional layers with recurrent ones. This is due to their ability to learn hierarchical representations of data and exploit temporal dependencies in inputs. There are also architectures that use meta-parameters such as attention vectors or word embeddings; however, they do not provide significant improvements over RNNs alone. We have also discussed the application of deep learning to facial emotion recognition, as well as its limitations. We conclude by discussing some future work that can be done in this field. The paper reviews the literature on face recognition. It also explains the relationship between facial emotion and facial feature extraction, which is essential for emotion recognition.