Enhancing local binary pattern – support vector machine (LBP-SVM) for improved facial emotion classification

ABSTRACT: Supervised Machine Learning algorithm like Support Vector Machines (SVMs), has emerged as powerful tools for classification, regression, and anomaly detection tasks. However, despite their high potential for facial emotion classification, SVMs face several challenges, including manual hyperparameter tuning, overfitting in handling overlapping classes and noisy features, and failure to effectively detect action units. Addressing these challenges is important in improving the accuracy and robustness of Facial Emotion Recognition (FER) systems. This study proposed improvements to the Local Binary Pattern – Support Vector Machine (LBP-SVM) framework by incorporating several key techniques. The Grid Search will be used to automatically select the optimal value for C and gamma parameters. For overfitting and noisy features, Recursive Feature Elimination will be used to select the features that contribute the most to class separability. Finally, to further enhance the accuracy of the model, the researchers also combined the features extracted from Facial Action Units to capture the dynamic characteristics of facial expressions. Experimental results on the CK+ datasets show that the enhanced LBP-SVM achieves a highest mean accuracy of 98.34% across 10-fold cross-validation, demonstrating improved robustness in recognizing subtle and complex emotional expressions.

Filipiniana