Accuracy Enhancement for Ear Acoustic Authentication Using Between-class Features

TL;DR

This paper introduces a novel between-class feature method for ear acoustic authentication, significantly reducing false rejection rates and improving accuracy without compromising security.

Contribution

It proposes a new feature generation technique and SVM-based learning approach to enhance ear biometric authentication accuracy.

Findings

FRR reduced by 7.95% at FAR=0.1%

Equal error rate decreased by 0.15%

Authentication accuracy improved with BC features

Abstract

In existing biometric authentication methods, the user must perform an authentication operation such as placing a finger in a scanner or facing a camera. With ear acoustic authentication, acoustic characteristics of the ear canal are used as biometric information. Therefore, a person wearing earphones does not need to perform any authentication operation. In biometric authentication, it is necessary to minimize the false acceptance rate (FAR) so that no unauthorized user is misidentified as an authorized user. However, if the FAR is set low, it increases the false rejection rate (FRR), the rate at which authorized users are falsely recognized as unauthorized users. It has been reported that when FAR is 0.1%, the FRR in ear acoustic authentication reaches as much as 22%. In this study, we propose a method that reduces FRR and enhances authentication accuracy; it generates new ear canal acoustic characteristics called between-class (BC) features, which combine the ear canal acoustic characteristics of authorized and unauthorized users features. The proposed method uses a support vector machine to learn the BC features as the data of authorized users, then generates a hyperplane in an area close to that data. We hypothesize that this would decrease the possibility of misidentifying an unauthorized user as an authorized user and decrease the FRR when the FAR is 0.1%. To evaluate the performance of the proposed method, BC features were applied to ear acoustic authentication, and FAR and FRR were calculated. The FRR with FAR = 0.1% was 7.95% lower than with the previous method, and the equal error rate -- the error rate when FAR and FRR are equivalent -- decreased by 0.15%. These results confirmed that the proposed method can make ear acoustic authentication more convenient while maintaining a high level of security.