Improving Gait Cryptosystem Security Using Gray Code Quantization and Linear Discriminant Analysis

Abstract

Gait has been considered as an efficient biometric trait for user authentication. Although there are some studies that address the task of securing gait templates/models in gait-based authentication systems, they do not take into account the low discriminability and high variation of gait data which significantly affects the security and practicality of the proposed systems. In this paper, we focus on addressing the aforementioned deficiencies in inertial-sensor based gait cryptosystem. Specifically, we leverage Linear Discrimination Analysis to enhance the discrimination of gait templates, and Gray code quantization to extract high discriminative and stable binary template. The experimental results on 38 different users showed that our proposed method significantly improve the performance and security of the gait cryptosystem. In particular, we achieved the False Acceptant Rate of 6×10^−5% (i.e., 1 fail in 16983 trials) and False Rejection Rate of 9.2% with 148-bit security.