Effective countermeasure against biometric spoofing attacks using unsupervised one-class learning

Vishu Gupta, Masakatsu Nishigaki, and Tetsushi Ohki
IWIN2018, September 2018. (Best Paper Award)


With the advent of new technologies, the methods of presentation attacks as well as the security measures taken against it is diversifying with each passing day and are competing with each other. Yet the imposter is able to make an access to a system illegally by deceiving the security through the use of material containing artificial biometrics traits like printed photo, display etc. Therefore, we propose a novel presentation attack detection algorithm which can ensure security against unknown presentation attacks without any prior knowledge of fake samples. Moreover, our proposed algorithm can detect presentation attack with single static image only. The key tasks are divided into two parts, creating a smooth manifold of live samples and determining whether the query image is included in the manifold. In this paper, we utilize one class system such as SVM(Support Vector Machine) and DCGAN(Deep Convolutional Generative Adversarial Network) to learn the manifold of live samples. For DCGAN we propose a liveness scoring scheme based on the AnoGAN(Anomaly Generative Adversarial Network) framework. Based on these, we utilize the proposed method to face presentation attack detection. Through our experiment we were able to produce decent results by using palm live/fake image dataset.