Search Results

Now showing 1 - 2 of 2
  • Item
    Colour Processing in Adversarial Attacks on Face Liveness Systems
    (The Eurographics Association, 2019) Abduh, Latifah; Ivrissimtzis, Ioannis; Vidal, Franck P. and Tam, Gary K. L. and Roberts, Jonathan C.
    In the context of face recognition systems, liveness test is a binary classification task aiming at distinguishing between input images that come from real people's faces and input images that come from photos or videos of those faces, and presented to the system's camera by an attacker. In this paper, we train the state-of-the-art, general purpose deep neural network ResNet for liveness testing, and measure the effect on its performance of adversarial attacks based on the manipulation of the saturation component of the imposter images. Our findings suggest that higher saturation values in the imposter images lead to a decrease in the network's performance. Next, we study the relationship between the proposed adversarial attacks and corresponding direct presentation attacks. Initial results on a small dataset of processed images which are then printed on paper or displayed on an LCD or a mobile phone screen, show that higher saturation values lead to higher values in the network's loss function, indicating that these colour manipulation techniques can indeed be converted into enhanced presentation attacks.
  • Item
    Training Dataset Construction for Anomaly Detection in Face Anti-spoofing
    (The Eurographics Association, 2021) Abduh, Latifah; Ivrissimtzis, Ioannis; Xu, Kai and Turner, Martin
    Anomaly detection, which is approaching the problem of face anti-spoofing as a one-class classification problem, is emerging as an increasingly popular alternative to the traditional approach of training binary classifiers on specialized anti-spoofing databases which contain both client and imposter samples. In this paper, we discuss the training protocols in the existing work on anomaly detection for face anti-spoofing, and note that they use images exclusively from specialized anti-spoofing databases, even though only common images of real faces are needed. In a proof-of-concept experiment, we demonstrate the potential benefits of adding in the anomaly detection training sets images from general face recognition, rather than specialised face anti-spoofing, databases, or images from the in-the-wild images. We train a convolutional autoencoder on real faces and compare the reconstruction error against a threshold to classify a face image as either client or imposter. Our results show that the inclusion in the training set of in-the-wild images increases the discriminating power of the classifier on an unseen database, as evidenced by an increase in the value of the Area Under the Curve.