Verbal language has become the main way to communicate our ideas since the already established linguistic signs allow understanding between people. However, this is not the only way we have to communicate, since nonverbal language, especially facial expressions, usually convey a lot of information. Currently, the use of Convolutional Neural Networks (CNN) has allowed us to identify these emotions more easily through facial expression recognition (FER), which has attracted much attention from various fields of research. In this work, we will provide detailed information about the currently most used dataset and methods for identifying emotions using facial expressions, such as VGG16, ResNet50 and Inception-V3, obtaining a better performance in ResNet-50. This survey shows that the main differences in precision in each architecture are due to the number of images in the datasets used for training and testing.
- automatic emotion recognition
- convolutional neural network
- deep learning
- facial expression recognition