Improving the generalization capability of face spoofing detection

• Main Author: Li, Haoliang
• Other Authors: Kot Chichung, Alex
• Format: Theses and Dissertations
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Biometrics
• Online Access: https://hdl.handle.net/10356/89790; http://hdl.handle.net/10220/46392
• Institution: Nanyang Technological University

Face spoofing detection has received great research interest recently due to the rapidly increasing demand in user authentication with facial information. The traditional face spoofing detection methods are developed based on either hand-crafted feature or deep learning based feature, replying on sufficient representative training data. However, these methods are limited in their scope that face images (videos) for training and testing are all collected from similar capture conditions, which limits their practical applications since the environment of face capturing can be diverse in real world. This thesis will present three different face anti-spoofing methods with improved generalization capability to new face capturing conditions and environment from three perspectives, where the training data can be scare, unlabelled and even no training data is available in the application specific domain. The corresponding learning strategies and face spoofing methods are further developed based on these practical application scenarios.