A research team led by Ri Jin Gwang, a researcher at the Faculty of Distance Education, has made a study of face anti-spoofing by spoof cues learning to give a solution to the problems arising in face authentication, one of the common problems in distance education.
Generally, typical face anti-spoofing methods used intrinsic features of attack mediums such as a printed photo or video replay, and selecting discriminative features between live samples and spoof ones became automatic with the development of deep learning. As a result, spoofing detection became a binary classification problem for distinguishing between false and true.
However, the existing methods might not get enough discriminative features or tend to overfit predefined datasets, which leads to some problems with generalization. Limited generalization capacity of FAS is attributable to the diversity of spoof samples including unknown ones.
Therefore, the research team has introduced spoof cues to improve the generalization capacity of a learning model and the correctness of FAS.
As the proposed method has been introduced for learner identification of distance education on the mobile network, face anti-spoofing caused by printed photos or video replay is not a problem any more. It means the practicability and scientific accuracy of online education are fully guaranteed.
...
