Deep Learning for Facial Expression Recognition : Advancing Artificial Intelligence in Emotional Understanding

Abstract

Facial expressions are a primary channel for non-verbal communication, making their automatic recognition a critical task in advancing human-computer interaction (HCI), robotics, and affective computing. While deep learning has shown promise, developing models that are both accurate and robust across diverse conditions remains a significant challenge. This study proposes a high-performance approach for Facial Expression Recognition (FER) by leveraging deep learning and transfer learning. Our method utilizes a transfer learning model based on the VGG16 architecture, which was selected for its proven success in complex image recognition tasks. The model was fine-tuned using transfer learning and implemented with the TensorFlow and Keras libraries. To validate its effectiveness, the system was rigorously evaluated on two standard benchmarks : the Cohn-Kanade (CK+) and the Facial Expression Recognition 2013 (FER2013) datasets. The proposed model achieved an outstanding accuracy of 96% on the CK+ dataset and a competitive accuracy of 69% on the more challenging, in-the-wild FER2013 dataset. These results demonstrate the efficacy of our transfer learning-based approach and highlight its potential for creating reliable and practical FER systems.