Analysis of FaceNet and VGG16 for Blind Face Recognition with MTCNN and HaarCascade Detection Methods

Face recognition is a rapidly growing biometric technology, especially with the application of Convolutional Neural Networks (CNN) such as FaceNet and VGG16. This research aims to evaluate the effectiveness of both CNN models in recognizing the faces of visually impaired people, who face the challenge of limited vision in image retrieval. The research uses two face detection methods, namely MTCNN and HaarCascade, to analyze the effect of face detection on recognition accuracy. The experimental method was conducted by collecting facial data of visually impaired people under various lighting conditions and expressions. The results show that accurate face detection greatly affects the performance of face recognition models, with MTCNN providing better face detection results (93.75% detection accuracy) than HaarCascade (83.75% detection accuracy). Both models, FaceNet and VGG16, show excellent face recognition accuracy (100%) if the face image is correctly detected by MTCNN. Therefore, for face recognition of visually impaired people, it is recommended to use MTCNN as the face detection method, followed by FaceNet or VGG16 for face recognition.