User authentication via finger-selfies

Abstract

In the last one decade, the usage and capabilities of smartphones have increased multifold. To keep data and devices secure, fingerprint and face recognition-based unlocking are gaining popularity. However, the additional cost of installing fingerprint sensors on smartphones questions the use of fingerprints. Alternatively, finger-selfie, an image of a person’s finger acquired using a built-in smartphone camera, can act as a cost-effective solution. Unlike capturing face selfies, capturing good-quality finger-selfies may not be a trivial task. The captured finger-selfie might incorporate several challenges such as illumination, in- and out-of-plane rotations, blur, and occlusion. Users may even present multiple fingers together in the same frame. In this chapter, we propose authentication using finger-selfies taken in an unconstrained environment. The research contributions include the UNconstrained FIngerphoTo (UNFIT) database which is captured under challenging unconstrained conditions. The database also contains the manual annotation of identities and location of the fingers. We further present a segmentation algorithm to segment finger regions and, finally, perform feature extraction and matching using CompCode and ResNet50. Experimental results show that despite multiple challenges present in the UNFIT database, the segmentation algorithm can segment and perform authentication using finger-selfies.