Research On 3D Face Recognition Based on Deep Learning Reconstruction

To address the problems of insufficient 3D face data, high acquisition costs and low recognition accuracy, this paper proposes a framework of 3D face recognition algorithm based on deep learning reconstruction. Firstly, two-dimensional face features are extracted by High Resolution Network (HRNet), and the reconstructed three-dimensional face is obtained by using Graph Convolutional Network (GCN). Feature matching is completed using the loss function proposed by Arcface to recognise the reconstructed 3D faces. Using 2D face recognition assistance for 3D face recognition, the similarity between the 2D face and the reconstructed 3D face is weighted to improve the problem of face recognition being affected by complex environments. The experimental results show that the recognition accuracy of the 3D face recognition algorithm based on deep learning reconstruction is 95.83%, and the recognition effect is better than that of 2D face recognition alone as well as 3D face recognition alone.