Online co-training in mobile ocular biometric recognition

Number of co-training based adaptation schemes have been proposed to address the issue of biometric template representativeness to the intra-class variation in changing environments. However, adaptation schemes for mobile environments remain limited. This is because mobile devices require online labeled data and efficient template management strategy due to limited hardware resources. The aim of this paper is to propose an online co-training scheme for mobile ocular biometrics. Experimental investigations on large scale VISOB dataset suggest a reduction in the Equal Error Rate (EER) of about 25.9% on adaptation due to co-training for ocular biometrics in mobile environment.