Bioelectrical Signals as Emerging Biometrics: Issues and Challenges

Table 1

Performance estimates of various methods using ECG as a biometric under different validation and test conditions. SIMCA: soft independent modeling of class analogy, NN: neural network, PCA: principle component analysis, LDA: linear discriminate analysis, EER: equal error rate, and N/R: not reported.

Methods	Data size	Data representation techniques	Classification techniques	Classification accuracy	Test conditions
Israel et al. [2]	29 subjects	Analysis of waveform curvature	Linear discriminant analysis	97-98%	Normal and different anxiety states
Irvine et al. [3]	43 subjects	Eigen pulse analysis, PCA	Correlation based analysis	~100% enrollment rate	Varying stress levels
Biel et al. [5]	20 subjects	Siemens ECG using Siemens Megacart	Multivariate analysis SIMCA	N/R	Validated ECG data recorded in 6 weeks
Shen et al. [6]	20 subjects	Feature extracted from QRST fiducials	Template matching, decision-based NN	80–100%	MIT-BIH database [29]
Wang et al. [7]	31 subjects	Autocorrelation and discrete cosine transform	Nearest neighbor classifier (NN)	94.47%–97.8%	PTB and MIT-BIH database [29]
Singh and Gupta [9, 10]	50 subjects 250 ECG recordings	Time derivative and waveform curvature analysis	Correlation based classification	98.09%; EER: 1.01%	Physionet database [29]
Multibiometric system using ECG signal Singh et al. [27]	78 subjects	Time derivative and waveform curvature analysis	Transformed based score fusion and classification	99.08%; EER: 0.2%	Physionet database [29], face and fingerprint NIST [33]