Robust multi-spectral imaging for ocular biometric

Abstract

Recent development of sensors has allowed the biometric community to explore the possibility of authentication beyond visible spectrum. More importantly, multi-spectral imaging in biometric has shown great potential to work robustly under unknown varying illumination conditions for face recognition. While face biometrics in traditional settings has also indicated the applicability of ocular region for improving the performance, there are not many works that have explored recent imaging methodologies. In this paper, we present the study that explores the possibility of recognizing ocular biometric feature using multi-spectral imaging. While exploring the possibility of recognizing the periocular region in different spectral bands, this work also presents the performance variation of periocular region for cross-spectral matching. We have captured a new ocular image database in eight narrow spectral bands across Visible (VIS) and Near-Infra-Red (NIR) spectrum (530nm to 1000nm) using our custom built sensor. The data consists of images from 52 subject with a sample size of 4160 spectral band images captured in two different sessions. The extensive set of experimental evaluation results obtained on the state-of-the-art methods indicates highest recognition rate of 96.92 percent at Rank -1, demonstrating the potential of multi-spectral imaging for robust periocular recognition.