Emotion Recognition from Facial Expressions using Deep Recurrent Attention Network

Facial expressions serve as a powerful nonverbal communication tool. They convey emotions effectively without the need for verbal expressions. Extensive research endeavors have been dedicated to this field, primarily focusing on employing Neural Networks (NNs) for feature extraction and inference. However, these studies exhibit substantial variation concerning NN architectures and other contributing factors. Hence, this paper introduces a Deep Convolutional Recurrent Attention Network (DCRAN) model architecture with notable performance improvement for Facial Emotion Recognition (FER). The proposed DCRAN model's effectiveness is demonstrated through validation using the well-known FER2013 dataset, achieving a test accuracy of 81.1%.