Learning-Based Index-of-Maximum Hashing for Touch-Stroke Template Protection
Jinwei Zhi, Ooi Shih Yin, Andrew Beng Jin Teoh · 2019
Touch-stroke dynamics is an emerging biometrics which captures a user's unique way of swiping a set of “strokes” on the mobile devices, and uses them as a matching template. Unexceptional to other biometrics, this template is also vulnerable to adversary attacks that may lead to privacy invasion and impersonation risk. One of the ideal ways to protect the biometric template is through the method of cancelable biometrics. On top of distorting the original template, it makes the template revocable as well. In this work, we propose an improved cancelable touch-stroke template method based on the Index-of-Maximum (IoM) hashing [10], and termed as learning based IoM (LIoM) hashing throughout this paper. Unlike vanilla data-agnostic IoM hashing, data-driven based LIoM hashing utilizes the supervising learning mechanism to generate a more discriminative and compact cancelable touch-stroke template. The proposed method achieved better performance than original template on Frank touch-stroke dynamic database with same feature length as well as satisfies three template protection design criteria, namely irreversibility, revocability and unlinkability.