Smartphone authentication system using periocular biometrics

The increasing usage of smartphones has raised security concerns regarding these devices due to presence of high amount of personal and sensitive data. The risk is higher without a proper mechanism to handle the authentication to access the smartphone device. In this work, we present a standalone modular biometric system based on periocular information to authenticate towards device. The proposed system has been implemented on the Android operating system. We field tested and evaluated the proposed system using a new database acquired capturing samples with three different devices. We apply the three well known feature extraction techniques, SIFT, SURF and BSIF independently in the proposed peroicular based authentication system. The best performance achieved with GMR = 89.38% at FMR = 0.01% indicates the applicability of the proposed periocular based mobile authentication system in a real-life scenario.

I. Introduction

Smartphones being a widely used personal device, are also used to store personal and sensitive data. Misuse of a smartphone due to low security to control the access to the device can lead to loss of personal data, which can be used to hack the accounts related to the owner of the device. Traditional methods have employed numeric, alphabetic or alpha-numeric PIN codes to secure the device, which are limited to fixed length. In consequence, the security of such access control methods is very limited, when expressed as entropy of the PIN code. For the most common case of arabic numerals (0–9), the symbol count is 10 and thus the entropy per digit H=3.322 bits. For a common 4-digit password the entropy is approximately 13 bits. Extending the password in length targeting at higher entropy reduces the usability of the method by creating hassle in managing multiple passwords of longer length [1]. Alongside the other problems, a simple brute force approach can be successful in cracking a short password. A plausible solution is to base the systems on biometric characteristics for authentication, which have higher entropy than passwords [2], [3].