Research of 3D face recognition algorithm based on deep learning stacked denoising autoencoder theory

• Published in: 2016 8th IEEE International Conference on Communication Software and Networks (ICCSN) pp. 663 - 667
• Main Authors: Jian Zhang, Zhenjie Hou, Zhuoran Wu, Yongkang Chen, Weikang Li
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2016
• Subjects: 3D face depth data; Analytical models; Data models; deep learning; Face; Face recognition; Machine learning; neural networks; stacked denoising autoencoder; Three-dimensional displays; Training; unsupervised preliminary training
• DOI: 10.1109/ICCSN.2016.7586606

This electronic Due to the fact that the 3D face depth data have more information, the 3D face recognition is attracting more and more attention in the machine learning area. Firstly, this paper selects 30 feature points from the 113 feature points of Candide-3 face model to characterize face, which improves the efficiency of recognition algorithm obviously without affecting the recognition accuracy. With the significant advantage of the characterization of essential features by learning a deep nonlinear network, this paper presents a stacked denoising autoencoder algorithm model based on deep learning which improves neural networks model. This algorithm conducts the unsupervised preliminary training of face depth data and the supervised training to fine-tuning the network which is better than neural network's random initialization. The experiment indicates that compared with real face data, the reconstruction face model has a small matching error by using SDAE algorithm and it achieves an excellent face recognition effect.