Optimization of facial recognition authentication system using InceptionResNetV1 with Pretrained VGGFACE2

Face recognition as a biometric authentication method continues to evolve due to its high security and ease of use. However, training models from scratch faces challenges such as the need for large datasets and high computational resources. This study aims to optimize the face authentication system using the InceptionResNetV1 architecture with a transfer learning approach from the pretrained VGGFace2 model and to compare its performance with CASIA-WebFace. Face detection is conducted using YOLOv8, face embeddings are generated by InceptionResNetV1, and authentication is performed by calculating the Euclidean distance between embeddings. Face data were collected from university students and divided into training and testing datasets. Performance evaluation includes accuracy, precision, recall, F1-score, and the confusion matrix. The results show that the VGGFace2 model achieved an accuracy of 98.75%, a recall of 100%, and an F1-score of 99.26%, with no False Negatives, while CASIA-WebFace achieved an accuracy of 86.25% with a recall of 85.07%. The main contribution of this study is to demonstrate that the use of transfer learning with the pretrained VGGFace2 model can significantly improve the accuracy of face authentication systems and to show its effectiveness for developing systems with limited data and computational resources. This study contributes by highlighting the superiority of the pretrained VGGFace2 model in face authentication systems and emphasizing the effectiveness of transfer learning for implementing accurate systems under resource constraints. Keywords: Authentication System, InceptionResNetV1, Face Recognition, Transfer Learning, VGGFace2