Biometrics are unique and specific data about an individual, enabling their identification and authentication. Face-based biometrics are the most stable in determining a person’s identity among various known biometrics. Moreover, face-based biometrics also have the potential to be applied in many fields, such as security, finance, and administration, leading to increased research in face recognition. Research on face recognition based on convolutional neural networks (CNN) has gained popularity since the architecture of AlexNet won the ImageNet competition in 2012. Since then, various CNN-based face recognition architectures have emerged, including VGGNet, GoogleNet, FaceNet, and ResNet.
A deep learning-based face recognition model named TURiMuka (Teknologi Unggul Rekognisi Muka or Advanced Face Recognition Technology in English) was created using the ResNet50 model, transferred-learning using the dataset of academic staff members at Universitas Syiah Kuala. The model achieved an F1 score of 89%. Several methods of facial variations were analyzed and tested, including geometric augmentation, facial movement in various yaw, pitch, and roll angles, and the HyperStyle facial synthesis. Subsequently, ResNet, MobileNetV3, and SeResNet architectures were trained using a combination of the original facial images and images with variations. The model with the best performance was employed for the TURiMuka prototype.
