Federated Multitask Learning with Manifold Regularization for Face Spoof Attack Detection

Figure 1

The flowchart of the proposed federated learning framework for face spoof attack detection. (a) For each subsystem for face recognition, we set up a task to detect face spoof attack. Facial images are represented by local binary pattern (LBP) features. These features are used to train local models of subsystems. (b) Local models are transferred to the server and multitask learning is applied to train the global model. (c) Finally, the global model is transferred back to subsystems and it can be used for face spoof attack detection on them.