Comparative Study on ECG Based Person Identification System Based on Non Uniform Filter Bank Features Vectors using Vector Quantization and GMM Modeling Techniques

Abstract

The objective of research work is to experimentally verify the significance of Electrocardiogram (ECG) signal in biometric person authentication. The motivation is based on the earlier work of demonstrating the feasibility of using Electroencephalogram (EEG) signal for person identification [1]. The ECG signals are processed by the cepstral analysis plus Vector Quantization techniques (VQ), Non uniform filter bank plus Vector Quantization and Non uniform filter bank plus Gaussian Mixture Modelling (GMM) to develop person-specific models. The testing of these models indeed found to be person-specific. This experimental work demonstrates that by using simple lead I ECG data recorded through the simple low cost hardware, with the help of traditional feature extraction techniques and widely using modelling techniques like VQ and GMM, it is possible to build the person-specific models for identification of persons. The ECG database of 49 healthy subjects indeed confirms this fact. The studies performed in this work indeed show a promising direction for using ECG as a biometric feature. The best score of person identification using different feature extraction methods along with modelling techniques using 49 persons database can be obtained if the training data include intermixing of three sessions of ECG data The results reveal that the NUFB plus its derivatives with VQ and GMM give highest performances of 93.87%, 97.95% respectively, compared to the cepstral plus its derivatives with VQ, whose highest performance is 91.83%.