Identifying Human Emotions from Facial Expressions with Deep Learning

Abstract

The identification of facial expressions that reveal human emotions can help computers to better assess the human state of mind, so as to provide a more customized interaction. We explore the recognition of human facial expressions through a deep learning approach using a Convolutional Neural Network (CNN) algorithm. The system uses a labelled data set containing around 32,298 images with multiple facial expressions for training and testing. The pre-training phase involves a face detection subsystem with noise removal, including feature extraction. The generated classification model used for prediction can identify seven emotions of the Facial Action Coding System (FACS). Results of our work in progress demonstrate an accuracy of 79.8% for the recognition of all basic seven human emotions, without the application of optimization techniques.