Enhancing facial expression recognition through generative adversarial networks-based augmentation

Emotion plays a significant role in our daily lives. It can describe the inner feelings and state of an individual and contribute to the communication process. Human–machine interaction is possible as a result of the application of these expressions. Facial expression recognition requires a significant amount of facial images as input data. However, such datasets pose challenges related to image quality and sample imbalance. Since facial expressions exhibit a high degree of diversity, accurately classifying them is a challenging task, particularly for expressions that have fewer samples. Building an efficient and reliable system requires a substantial amount of data. This study aims to address the issue of class imbalance in facial expression datasets by developing and implementing a deep learning-based classification model that uses synthetic images generated through Generative Adversarial Networks. The goal is to improve recognition accuracy for each expression. The effectiveness of the proposed augmentation technique is compared with simple augmentation techniques using VGG16 and the proposed DCNN Model. GAN-based augmentation and the proposed deep learning model outperformed by a large margin on the FER-2013 dataset.