Fitra Abdurrachman Bachtiar, Mufidatun Nuha
Feature extraction is a critical component of facial expression recognition, requiring careful selection to ensure optimal performance. Traditional template-based models often struggle with computational efficiency and are sensitive to factors such as illumination, color variation, and image quality. Although geometry-based models have demonstrated robustness in feature extraction, further optimization remains needed to enhance the recognition accuracy. In this study, the CK+ dataset is utilized, and employ facial landmarks as a geometry-based feature extraction method. The extracted features are then processed using an Extreme Learning Machine combined with Genetic Algorithms (ELM-GA) to select a subset of features and classify facial expressions. The proposed model effectively reduces the number of features used in the classification model by more than half, improving computational efficiency. Additionally, the accuracy of the proposed model outperforms the VGGNet and combination of machine learning and Local Binary Patterns (LBP) and machine learning and Gray-Level Co-occurrence Matrix (GLCM) for feature extraction. Although the current model does not represent the best possible solution, the promising accuracy results provide a solid foundation for further refinement and development in facial expression recognition systems. © 2024 IEEE.
Universitas Brawijaya, Intelligent System Laboratory, Computer Science Faculty, Malang, Indonesia