Date of Publication :15th March 2025

Abstract: A significant issue in affective computing is the identification of facial expressions, which influences behavioral analysis, mental health evaluation, and human-computer interaction applications. The issue of weak generalization emerges from conventional deep learning models' inability to effectively handle temporal and spatial relationships. The sophisticated framework for facial emotion recognition presented in this study is based on three distinct architectures: (1) a CNN structure augmente d with advanced feature extraction and regularization methods; (2) a CNN-LSTM combined model intended to capture sequential patterns in facial expressions; and (3) a refined CNN that emphasizes precise spatial feature extraction. To achieve optimal classification accuracy, robustness, and generalization, each model is subjected to individual tuning. Experiments performed on standard datasets reveal significant enhancements in accuracy, recall, and precision when compared to traditional techniques.Findings indicate that more complex models and consideration of sequence in modeling significantly improve facial emotion recognition, paving the way for more resilient and instantaneous affective computing.

Reference :