Inter-Device Periocular Recognition Under Near-Infrared Light

Abstract

Abstract Periocular biometrics is a relatively new field of research, and only several publications on this topic can be found in the literature. It can become a promising feature that can be used independently or as a complement to other biometrics. In this work, the recognition rates of periocular biometrics on a single acquisition device and inter-device database is verified and the impact of different image sources on the performance of recognition algorithms is investigated. For this purpose a NearInfrared Light database was collected. The database contains images taken by two acquisition devices. In order to test the periocular biometric trait, three feature extraction methods are chosen: Histograms of Oriented Gradients, Local Binary Patterns and Scale Invariant Feature Transform. The fusion of these methods is also proposed and it is tested on inter-device database. The feasibility of applying periocular recognition as an individual decision module for a biometric system is assessed. Experimental results yield Equal Error Rate of 17.65 for right eye using inter-device database of 640 gallery periocular images for each eye side taken from 32 different individuals (20 images per individual for each eye side). These results are obtained by the optimal weighted sum fusion of the three feature extraction methods.