Single Pulse ECG-based Small Scale User Authentication using Guided Filtering

Abstract

Electrocardiogram (ECG) has been demonstrated as a promising biometric for user authentication or classification. However, most of the previous works on ECG biometrics dealt with more than five ECG pulses at once, which will require at least a few seconds to acquire. Moreover, many of them investigated classification systems that require ECG signals of many people for effective dimensionality reduction methods such as PCA or for powerful classifiers such as SVM. In this article, we consider fast responding, small scale authentication systems (e.g., wearable devices). We investigate the feasibility of using a single pulse ECG for authentication assuming that there is no access to others’ ECG signals. Multiple ECG pulses are allowed only in the enrollment stage. We propose to use guided filter (GF) to reduce noise of a single ECG pulse using a low noise ECG template from the enrollment step. We employed simple distance measures such as Euclidean distance and dynamic time warping (DTW) for small scale authentication system and compared them with PCA based authentication system using others’ ECG information. We evaluated our proposed methods with public ECGID database (89 subjects, selected 2 records per subject that were collected on the same day). Performance measures were used such as inter/intra distance ratio (IIDR), area under the curve (AUC) of the receiver operating characteristic (ROC) curve, and equal error rate (EER). GF improved the performance of simple user authentication systems with Euclidean distance and DTW substantially. The Euclidean distance with GF achieved comparable authentication performance with PCA based method using other people’s ECG information (EER = 2.4%).