Broad Recognition of Fake Faces

Deep learning system (DLS) always require a time-consuming process because of too many hyperparameters and complex structure. When we update some data in DLS, a complete retraining process was encountered which spending too much time. To get over these shortcomings, the Broad Learning System (BLS) is established by C.L.Philip Chen, which aims to make a different way of artificial intelligence. By the BLS, the origin inputs are transformed and the structure of the system expanded in the feature nodes (The following is expressed as fn) and enhancement nodes (The following is expressed as en) in the wide sense. In this paper, we apply the BLS to the recognition of fake faces. Differing from previous studies, we further validate the anti-interference ability of BLS by involving a perturbation matrix P, which does not change incremental learning processes of fn and en. According to the performance on the considered dataset, BLS is more effective than DLS in recognition of fake faces. Further validation on other datasets are still necessary.