Rapid advancement in sensor technology through miniaturization of electronic components has enabled the consumer electronic (CE) research community including the manufacturers to embed various utility sensors into handheld devices. In addition to traditional sensors such as Inertial Measurement Unit (IMU), camera, fingerprint or proximity, futuristic sensors such as Electroencephalogram (EEG) or Electromyography (EMG) are also being included in the next-generation CE devices. Air or touch signature-based authentication systems are common in modern CE devices. However, cerebral activities clubbed with gestures will certainly enhance the security of such authentication systems. This can help consumers from being the victims of shoulder surfing attacks. In this article, a new method is proposed to verify air signatures by analyzing finger movements and cerebral activities together with the help of sensors in next-generation CE devices. Signatures are first spotted by analyzing 3D geometrical features of the finger movement during the signing. Concurrent EEG responses are then analyzed for the verification. Hidden Markov Model (HMM) and Random Forest (RF) classifiers have been used to train the system. Experiments reveal that EEG signals are highly correlated with the finger movements during air signatures even in the presence of motion artifacts. Therefore, false-positive rates have significantly reduced as compared to the existing tracking-based methods. Verification accuracy as high as 95.5% (HMM) and 98.5% (RF) have been recorded when tested on our dataset.

中文翻译：

---

通过智能地结合3D手指运动和脑部反应的手持电子设备的强大生物特征认证系统

通过电子组件的小型化，传感器技术的快速发展使包括制造商在内的消费电子（CE）研究界能够将各种实用传感器嵌入到手持设备中。除了惯性测量单元（IMU），相机，指纹或接近度之类的传统传感器外，下一代CE设备中还包括诸如脑电图（EEG）或肌电图（EMG）之类的未来派传感器。基于空中或触摸签名的身份验证系统在现代CE设备中很常见。但是，以手势为中心的大脑活动肯定会增强这种身份验证系统的安全性。这可以帮助消费者避免遭受肩膀冲浪攻击。在本文中，提出了一种通过在下一代CE设备中借助传感器分析手指运动和大脑活动来验证空气信号的新方法。首先通过分析签名过程中手指运动的3D几何特征来发现签名。然后分析并发的EEG响应以进行验证。隐马尔可夫模型（HMM）和随机森林（RF）分类器已用于训练系统。实验表明，即使在存在运动伪影的情况下，脑电信号在空中信号传递过程中也与手指运动高度相关。因此，与现有的基于跟踪的方法相比，假阳性率已大大降低。在我们的数据集上进行测试时，已记录出高达95.5％（HMM）和98.5％（RF）的验证准确性。