ABSTRACT
In recent years bioelectric signals like electrocardiogram, electroencephalogram, and photoplethysmogram based person authentication has received tremendous attention of the biometric community. This is due to the inherent advantages offered by these traits in comparison to the traditional biometrics like fingerprint, face, and iris. The unique qualities possessed by these traits ensure entire population coverage and alleviate the need of vitality check. An overview of biometrics with emphasis on ECG-based biometric recognition has been presented in this chapter. One non-fiducial feature extraction approach, which uses combination of autocorrelation and discrete cosine transform, has been discussed. The classification task has been carried out by employing two neural network architectures, namely, multilayer perceptron and radial basis function. Another set of experiments have been carried out using support vector machine as classifier in the multi-class classification mode. Two set of datasets have been used in these experiments, first one is Lead II ECG database and second dataset is Lead I database. Both the datasets have a population size of more than 100 subjects with multiple session recordings having sufficient gap between them. The experiments have been conducted for both within and across-session settings. The results obtained depict the efficacy of ECG as a trait which can be utilised for biometric recognition especially in the multimodal mode.
