Few-data guided learning upon end-to-end point cloud network for 3D face recognition

Abstract

Deep-learning-based 3D face recognition methods have developed vigorously in recent years, while the potential of these methods is being exploited in more and more scenarios. In this paper, an end-to-end deep learning network entitled Sur3dNet-Face for point-cloud-based 3D face recognition is proposed. The method uses PointNet, which is a successful point cloud classification solution but performs unexpectedly in face recognition, as the backbone. To adapt the backbone to 3D face recognition, modifications in network architecture and a few-data guided learning framework based on Gaussian process morphable model is supplemented. Instead of mass data in multiple datasets for training, our method takes only Spring2003 subset of FRGC v2.0 for training which contains 943 facial scans and the network is well trained with such a small amount of real data. The processing time to generate face representation is less than 0.15 s. Without fine-tuning on the test set, the Rank-1 Recognition Rate (RR1) is achieved as follows: 98.85% on FRGC v2.0 dataset and 99.33% on Bosphorus dataset, which proves the effectiveness and the potentiality of our method. When facing scenarios with limited resource, the proposed method is expected to give a competitive performance.