Face presantation attack detection by deep learning

Today, with the development of technology, artificial intelligence algorithms have started to be used in many fields. Artificial intelligence algorithms, which we use in many fields such as medical, weather forecasting, digital assistants, autonomous vehicles and security systems, are becoming more widespread and spreading to different fields day by day. Biometric recognition systems, which are the subject of this study, are actively used in many areas such as banking, health, airports and border control, and mobile phones. This study investigates the effectiveness of deep learning algorithms in face presentation attack detection. Attackers who threaten the security of facial recognition systems try to bypass these systems by using fake faces (e.g. photos, videos or masks). The aim of this study is to examine the success of deep learning-based methods in detecting such attacks. In this study, two different convolutional neural network models are used and different methods are added to these models to compare the problem solving performance of convolutional neural network models. The models are trained on the OULU-NPU dataset and the results obtained by using different data sampling methods on the dataset are discussed. The deep learning models, dataset and methodologies used in the thesis are described in detail, and the model training and testing processes are discussed in detail. For performance evaluation, the results obtained in the light of Half of Total Error Rate (HTER), Attack Presentation Classification Error Rate (APCER) and Bonafide Presentation Classification Error Rate (BPCER) metrics are discussed. The results show that the models are successful in detecting face presentation attacks. In conclusion, this thesis demonstrates that deep learning techniques are a powerful tool for face-presentation attack detection and can play an important role in improving the reliability of biometric security systems.