Facial Expression Recognition Using Deep Learning and Neural Embeddings

This study investigates Facial Expression Recognition (FER) as essential for understanding human emotions conveyed through facial expressions, involving face detection, facial expression detection, and classification.Recent advancements in deep learning have significantly enhanced FER accuracy, exemplified by combining Visual Geometry Group (VGG) and U-Net segmentation layers, achieving a remarkable 75.97% accuracy.Building upon prior research on neural embeddings, this study explores their application in improving FER models, focusing on basic models like VGG-19 and employing triplet loss.Extracted features are classified using various methods such as Support Vector Machine, XGBoost, Random Forest, and Artificial Neural Networks, with evaluation metrics including accuracy, precision, recall, and F1 Score.Findings indicate that modifications to the VGG19 classifier improve accuracy, with XGBoost attaining the highest accuracy of 65.70%.However, integrating triplet loss does not yield significant improvement, recording a highest accuracy of 65.30% when combined with the XGBoost model.These results suggest potential limitations, such as incorrect distance calculation methods and dataset imbalance, which need addressing for enhancing model efficacy and real-world applicability.Therefore, future research should focus on refining distance calculation techniques and ensuring dataset balance.