Heart‐ID: human identity recognition using heart sounds based on modifying mel‐frequency cepstral features. Issue 4 (29th June 2016)

• Record Type: Journal Article
• Authors: Abbas, Sherif N.; Abo–Zahhad, Mohammed; Ahmed, Sabah M.; Farrag, Mohammed
• Is Part Of: IET biometrics. Volume 5:Issue 4(2016)
• Page Start: 284
• Page End: 296
• Publication Date: 2016-06-29
• Subjects: cardiology -- phonocardiography -- cepstral analysis -- feature extraction -- biometrics (access control) -- channel bank filters -- nonlinear filters -- signal denoising -- signal classification -- medical signal processing; Heart‐ID -- human identity recognition -- mel‐frequency cepstral features -- heart sound signals -- phono‐cardio‐gram -- linear frequency cepstral features -- mel‐frequency equation -- triangular filters -- PCG signal -- wavelet packet filters -- nonlinear filter bank structure -- wavelet packet decomposition -- discriminant feature extraction -- wavelet denoising -- classification; Biometric identification -- Periodicals; 570.15195
• DOI: 10.1049/iet-bmt.2015.0033
• Languages: English
• ISSNs: 2047-4938

Abstract

Abstract : This study presents a new framework for human identity recognition using heart sound signals. The proposed framework is based on extracting cepstral features from heart sound signals, which are known as phono‐cardio‐gram (PCG). Two well‐known cepstral features have been adopted in most of the previously implemented PCG biometric authentication systems; namely, mel‐frequency and linear frequency cepstral features. In this study, two more cepstral features are proposed based on modifying the mel‐frequency cepstral features. The first one is based on modifying the mel‐frequency equation to increase the non‐linearity of the triangular filters in the frequency range of the PCG signal. The other is based on replacing mel‐scaled triangular filters with wavelet packet filters where a non‐linear filter bank structure is designed using wavelet packet decomposition to select the appropriate bases for extracting discriminant features. The proposed system uses wavelet de‐noising for pre‐processing and linear discriminant analysis for classification. The proposed system is evaluated on two databases; one consists of 21 users (BioSec. database) and the other consists of 206 users (HSCT‐11 database). Moreover, the proposed system is compared with previous systems that used the same databases. On the basis of the achieved results over the two databases, the two proposed cepstral features achieved higher correct recognition rates and lower error rates in identification and verification modes, respectively.