Optimised facial expression recognition via a hybrid ensemble classifier and improved feature extraction techniques

Facial expression recognition (FER) is crucial for applications in human-computer interaction, security, and healthcare. Traditional methods often struggle with issues such as limited generalisation, high computational demands, and handling complex emotions. To address these challenges, the research offers a novel methodology that combines advanced image pre-processing, feature extraction, and classification techniques. This initiates with image pre-processing that standardises facial images by reducing noise. Then the feature extraction employs a shallow convolutional neural network (CNN) for texture-based features. Then, the Softmax classifier for emotion prediction and incorporates hyperparameter tuning using the equilibrium optimiser (EO) algorithm. Additionally, feature selection is optimised using the improved cat swarm optimisation (ICSO) algorithm to refine the feature set. The integration of an ensemble classifier combining multi-layer perceptrons (MLPs) and ordinal logistic regression improves the robustness of the model, utilising soft labels. Experimental results demonstrate significant improvements over existing methods and achieve an accuracy of 99.12%.