User Re-Identification Using Clothing Information for Smartphones

With the expected of 2.5 billion people using smartphone by 2019, mobile biometric is a crucial field providing convenient and secure access for users. As most of the security systems only require single authentication, an intruder can gain access after the initial login stage. Human re-identification in mobile device is the task of continuously authenticating the person after initial login. In this paper, we investigate using clothing information for subject re-identification in mobile device. To this aim, we employed two approaches to extract shallow and deep clothing features, followed by a linear support vector machine (SVM) to distinguish genuine from impostors. Each method achieved promising results, with equal error rates below 0.05. Further, a fusion of handcrafted shallow features and data-driven deep features at, feature and score levels, provided 0.034 and 0.032 at EER, confirming the viability of the proposed method for accurate short-term re-identification for mobile use cases.