AN ENHANCED FACIAL EMOTION RECOGNITION FRAMEWORK USING GATED SPATIAL-FREQUENCY FUSION AND ENSEMBLE LEARNING

Facial expressions reflect a person's internal emotional state and are interpreted through automated facial emotion recognition (FER) methods in applications such as human-computer interaction, security, and healthcare. Early FER approaches relied on manual feature extraction combined with machine learning (ML) algorithms; however, such methods struggle with large-scale datasets and fail to capture the non-linear characteristics of facial expressions. To address these limitations, this study proposes an enhanced automated facial emotion classification framework based on an ensemble learning approach that evaluates the fused features at spatial and frequency levels. The spatial level features are extracted using the MobileNetV3 architecture, and the frequency level using the Fast Fourier Transform (FFT) method. Thereafter, these feature sets are integrated using the gated fusion method. The fused features are used for emotion classification by combining the predictions of ML algorithms (Random Forest (RF) and Support Vector Machine (SVM)) using a weighted voting classifier. Furthermore, hyper-parameter tuning of the ensemble learning approach using the Optuna optimizer enhances the performance of the proposed framework. According to the simulation evaluation, the proposed framework achieves an accuracy value of 0.7457 for the FER2013+ dataset and outperforms the previous approaches.