Smart Face Recognition System Using Hybridization of MTCNN and Facenet Algorithms

Face detection is a subfield of image processing that involves the identification of human faces from images or videos. Some common applications include facial recognition, emotion analysis, surveillance, and auto-tagging in photographs. Existing approaches suffer from less accuracy and high error. Therefore, a system with accuracy and precision is required to detect and identify the faces in real-time for security verification and surveillance. This paper aims to develop a face detection and recognition system. In this work, the Multi-task Cascaded Conventional Networks (MTCNN) algorithm is employed for face detection. It is a three-step convolutional network method, which is fast and accurate in detecting faces and obtaining the location of facial landmarks. Then, the FaceNet model is employed for face recognition. Once the faces are detected, the FaceNet model maps each face to a 128 vector, mostly preserving the distinct features of the face. These embeddings are further matched with a previously generated database to verify or find the person's possible identity. Extensive simulations have been conducted for performance analysis, and it was identified that the presented approach performs with improved accuracy compared to the peer methods.