Face Anti-Spoofing via Sample Learning Based Recurrent Neural Network (RNN)

Face biometric systems are vulnerable to spoofing attacks because of criminals who are developing different techniques such as print attack, replay attack, 3D mask attack, etc. to easily fool the face recognition systems. To improve the security measures of biometric systems, we propose a simple and effective architecture called sample learning based recurrent neural network (SLRNN). The proposed sample learning is based on sparse filtering which is applied for augmenting the features by leveraging Residual Networks (ResNet). The augmented features form as a sequence, which are fed into a Long Short-Term Memory (LSTM) network for constructing the final representation. We show that for face anti-spoofing task, incorporating sample learning into recurrent structures learn more meaningful representations to LSTM with much fewer model parameters. Experimental studies on MSU and CASIA dataset demonstrate that the proposed SLRNN has a superior performance than state-of-the-art methods used now.