With the wide use of personal consumer electronics devices such as smartphones, people store sensitive and confidential information more on their devices. Active authentication (AA) systems continuously authenticate users to reduce possible attacks after a successful login on the device. In this article, we propose match-on-card (MOC) approach for a secure active authentication scheme using touchscreen for smartphones to enhance the security and privacy and decrease the performance overhead on the consumer device. We train a Deep Neural Network (DNN) model, and store the model on the smart card available on the device for user authentication. To implement the user verification on smart cards, we quantize inputs to the model and the model’s parameters. A speed-up technique is added to the verification phase to improve the execution time. Evaluation results show that with a well configured DNN model, our on-card authentication reaches an Equal Error Rate (EER) of 2.6% for 15 strokes and verification time of 0.65 second for each stroke. Considering the average user’s stroke frequency of 1 stroke/s, our proposed scheme shows the potential for mobile MOC active authentication using touchscreen gestures on consumer devices.

中文翻译：

---

使用触摸屏生物识别卡主动认证的移动匹配

随着智能手机等个人消费电子设备的广泛使用，人们将更多的敏感和机密信息存储在他们的设备上。主动身份验证 (AA) 系统会持续对用户进行身份验证，以减少成功登录设备后可能发生的攻击。在本文中，我们为使用智能手机触摸屏的安全主动身份验证方案提出卡上匹配 (MOC) 方法，以增强安全性和隐私性并降低消费设备的性能开销。我们训练一个深度神经网络 (DNN) 模型，并将模型存储在设备上可用的智能卡上以进行用户身份验证。为了在智能卡上实现用户验证，我们量化了模型的输入和模型的参数。在验证阶段添加了加速技术以改善执行时间。评估结果表明，通过配置良好的 DNN 模型，我们的卡上身份验证在 15 次笔画中达到 2.6% 的等错误率 (EER)，每次笔画的验证时间为 0.65 秒。考虑到用户的平均笔画频率为 1 笔/秒，我们提出的方案显示了在消费设备上使用触摸屏手势进行移动 MOC 主动身份验证的潜力。