Biometrics from heart sounds: Evaluation of a new approach based on wavelet packet cepstral features using HSCT-11 database

Abstract

This paper introduces a new approach for human recognition using heart sounds. The main contribution of this paper involves adopting wavelet packet cepstral coefficients as new features for heart sound signals in biometric applications. The proposed features utilize a non-linear wavelet packet filter banks which are designed to match the acoustic nature of the heart sound. The proposed system is evaluated using an open database for heart sounds known as HSCT-11 which contains data collected from 206 users. Based on the achieved results, the proposed system can identify users with best accuracy of 91.05% and verify them with an equal error rate of 3.2% using 200-fold random validation (random sub-sampling). The experimental results showed higher correct recognition rates and lower error rates in identification and verification modes, respectively, compared to previously implemented systems evaluated on the same database (HSCT-11).