Compressed-Domain ECG-based Biometric User Identification Using Task-Driven Dictionary Learning

In recent years, user identification has become crucial for authorized machine access. Electrocardiography (ECG) is a new and rising biometrics signature. Rather than traditional biological traits, ECG cannot be easily imitated. In the long-term monitoring system, the wireless wearable ECG biomedical sensor nodes are resource-limited. Recently, compressive sensing (CS) technology is extensively applied to reduce the power of data transmission and acquisition. The prior CS-based reconstruction process aims at improving energy efficiency with different schemes, and they focus on the performance of reconstruction only. Therefore, we present a sparse coding-based classifier, trained by task-driven dictionary learning (TDDL), to realize low-complexity user identification in compressed-domain directly. TDDL is one of the dictionary learning and designed for classification tasks. It co-optimizes the dictionary and classifier weighting simultaneously, which gives better accuracy. In this article, we are proposing a TDDL-based compression learning algorithm for ECG biometric user identification as this directly identifies user identity (ID) without undergoing reconstruction process and conventional classifier. It can extract necessary information from the compressed-ECG signal directly to save the system power and computational complexity. The algorithm has 2%–10% accuracy improvements compared with state-of-the-art algorithms and maintains low complexity at the same time. Our proposed TDDL-CL will be the better choice in the long-term wearable ECG biometric devices.