A Novel Face Recognition Using Specific Values from Deep Neural Network-based Landmarks

The detection of facial landmarks has been an ongoing research topic for the past decade as it is used in facial recognition, facial expression analysis, and security purposes. This paper proposes a new face recognition algorithm that uses deep learning-based facial landmark detection algorithms to extract key features from images. Using the landmarks obtained from the applied deep neural network, three different features are extracted using specific values and used for face recognition. The use of specific values (SV) for facial recognition is the novelty of this work. Three specific values namely the cosine distance, angles and areas are derived from the coordinates of the landmarks. The recognition rates of faces using the extracted landmarks and SVs as the proposed features are evaluated through several experiments. Of the proposed features, the area provides the best result. Also, to investigate the effect of increasing the number of landmark points on the proposed face recognition rate, the MediaPipe face mesh algorithm is utilized. With the same chosen SVs, the recognition rate results are discussed when recognition was carried out with the increased number of landmarks.