On the vulnerability of ECG verification to online presentation attacks

Electrocardiogram (ECG) has long been regarded as a biometric modality which is impractical to copy, clone, or spoof. However, it was recently shown that an ECG signal can be replayed from arbitrary waveform generators, computer sound cards, or off-the-shelf audio players. In this paper, we develop a novel presentation attack where a short template of the victim's ECG is captured by an attacker and used to map the attacker's ECG into the victim's, which can then be provided to the sensor using one of the above sources. Our approach involves exploiting ECG models, characterizing the differences between ECG signals, and developing mapping functions that transform any ECG into one that closely matches an authentic user's ECG. Our proposed approach, which can operate online or on-the-fly, is compared with a more ideal offline scenario where the attacker has more time and resources. In our experiments, the offline approach achieves average success rates of 97.43% and 94.17% for non-fiducial and fiducial based ECG authentication. In the online scenario, the performance is de-graded by 5.65% for non-fiducial based authentication, but is nearly unaffected for fiducial authentication.

1. Introduction

As the Internet of things (IoT) becomes more popular in consumer, business, and military settings, one can expect the demand for biometric technologies to grow. IoT devices are supposed to outnumber the world's population this year [1] and their number should continue to dramatically increase for many years to come. Managing so many devices with passwords alone is ripe with challenges. In addition, the sensitive data gathered and stored by IoT could pose significant privacy concerns. Compared with conventional authentication techniques, such as digital passwords, personal identification numbers, and smart-cards/tokens, biometrics provide a more robust method for identifying a person, i.e., based on their distinctive physical characteristics. Biometrics can also be consid- ered more seamless and convenient, especially for continuous authentication [15]. That being said, it has already been demonstrated that many of the most popular biometric modalities (iris, face, fingerprint, and speech) can be spoofed and are, therefore, vulnerable to presentation attacks [27,12,13,9,3,6,28,25,24,11]. Over the past decade, alternative modalities based on biological signals have been explored and their resistance to presentation attacks is often highlighted as a major attribute. Notable examples include electrocardiogram (ECG) [8], photoplethys-mogram (PPG) [16], and electroencephalogram (EEG) [19], which possess high distinctiveness, are difficult to replicate, and provide intrinsic liveness detection. Among them, ECG has received the most attention and is beginning to gain larger acceptance from the biometrics community. For instance, ECG-based authentication systems, such as the Nymi wristband [2], are already coming to the market. ECG is a recording of the electric potential, generated by the electric activity of the heart, on the surface of the thorax that represents the extra cellular electric behavior of the cardiac muscle tissue. A typical, healthy ECG signal with different beats is shown in Figure 1. Generally speaking, ECG authentication systems can be categorized based on the feature extraction method (fiducial point vs. non-fiducial point) as well as the type of template matching used for classification. Fiducial point feature extraction relies on an accurate detection of ECG fiducial characteristic points such as P, Q, R, S, and T waves as shown in Figure 1, in order to obtain their relative amplitude, temporal intervals and morphological features. Non-fiducial point feature extraction analyzes an ECG in a holistic manner, typically by applying time or frequency analysis to obtain other statistical features. Despite the interest in ECG-based authentication, it's worth noting that ECG suffers from various noise sources such as motion, electromyography (EMG), and exercise, which can impact authentication accuracy [17]. In the literature, accuracy lies in the range of 94.3% to 100% [21]. Figure 1.

Waterfall plot of ECG beats collected from the same subject and localization of fiducial points.