COMPARATIVE ANALYSIS OF THE PERFORMANCE OF FACENET AND EIGENFACE MODELS WITH SUPPORT VECTOR MACHINE CLASSIFICATION UNDER LIGHTING VARIATIONS AND OCCLUSION CONDITIONS IN FACE RECOGNITION SYSTEMS

• Main Author: Risalah Ghaida, Dhiya
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82234
• Institution: Institut Teknologi Bandung

Emotional interaction between teachers and learners plays a crucial role in the educational environment. In an effort to support teachers to monitor and respond to each learner's emotion more effectively, this research addresses a suitable face recognition system method to be applied in a learner emotion recognition system. The research was conducted by comparing two models: FaceNet with Support Vector Machine (SVM) and Eigenface with SVM using a single sample per learner. The face detection process in this study uses the Multi-task Cascaded Convolutional Networks (MTCNN) method due to its reliability in detecting facial landmarks. The test results show the significant superiority of FaceNet and SVM models that achieve almost perfect accuracy of 98%, compared to the Eigenface model that only reaches 84%. Furthermore, under various lighting conditions, the FaceNet and SVM models still performed well with 90% accuracy, while Eigenface only achieved 9%. Under the partially covered face condition, FaceNet and SVM recorded 67% accuracy, much higher compared to 11% for Eigenface and SVM. The adaptability of FaceNet under various lighting conditions and visual challenges confirms that this deep learning model is very suitable for emotion recognition applications in online learning. Based on these results, this study recommends the use of FaceNet in learner emotion recognition systems to improve interaction quality and learning effectiveness.