Deep Learning Based Face Recognition System with Modified MTCNN and FaceNet

Nowadays face recognition is a good option for using as a bio-metrics for human identification and recognition system. This technology is used in various areas like attendance, security, payment, access control, etc., in a day to day life. The improvements in Face recognition models are increasing rapidly by using advanced deep learning techniques for large training datasets. However, face images have large variations like poses, resolutions, occlusions which are creating great challenges in face recognition related applications. In this work, we propose an advanced face recognition system that leverages a Multi-task Cascaded Convolutional Neural Network (MTCNN) and a deep neural algorithm called FaceNet. This combination of deep learning framework gave higher recognition accuracy. The proposed face recognition system utilizes the MTCNN model, comprising three cascaded networks (P-Net, R-Net, and O-Net), for accurate face detection and alignment. Subsequently, the FaceNet model extracts feature vectors, employing the Triplet Loss function to learn discriminative embeddings by minimizing the Euclidean distances between anchor-positive pairs and maximizing them for anchor-negative pairs. We evaluate our system on three benchmark datasets: LFW, Political Faces, and Celebrity Faces, achieving a high accuracy of 94.4% on the challenging LFW dataset.