Design Of Protected Iris Recognition Systems With Improved Authentication Performance

Chai, Tong Yuen (2022) Design Of Protected Iris Recognition Systems With Improved Authentication Performance. PhD thesis, UTAR.

Preview

PDF Download (3311Kb) | Preview

Abstract

Iris is unique with higher confidence in matching as compared to other biometric traits. No physical contact, difficult to spoof and not easily replaceable are among the advantages of iris recognition system. However, permanent identity loss will be experienced if the raw iris code is being stolen. Biometric template protection (BTP) schemes are implemented to increase public confidence in biometric systems regarding data privacy and security. The design of BTP has naturally incurred a loss of information which leads to performance degradation at the matching stage. Despite some extended works from these iris BTP schemes, there is still lack of a generalized solution for this problem. A trainable model that requires no further modification on the protected iris templates has been proposed in this thesis to improve the authentication performance. Improvement between 14% - 82% has been reported against four publicly available iris research databases: CASIAv1, CASIAv3, CASIAv4 and ND0405. Another BTP technique, namely key binding scheme can become vulnerable due to the inherent dependency of biometric features and the capacity of error correction code (ECC). Previous literature has shown deterioration in performance without the alignment process at iris codes. In this thesis, an alignment free cancelable iris key binding scheme without ECC has been proposed. The highest genuine acceptance rate (GAR) of 96.37% at zero false acceptance rate (FAR) has been achieved on CASIA-v3-interval iris database. The best performance can also be preserved with key length up to 200 bits. Next, focus has been put on the security concern caused by the iris code’s alignment process during matching. This process can only be conducted on probe iris codes for most of the protected iris recognition systems. In addition, high correlation between adjacent iris codes has indicated dependency along vertical direction. Thus, an iris template transformation method and a matching strategy are proposed to mitigate the alignment and inherent dependency issues of iris codes in this work. The proposed model has optimized the authentication performance of iris code by achieving EER as low as 0.46% on CASIA-v3interval iris database. Thus, vertical dependency in iris code has been mitigated while bit-shifting alignment can be applied directly onto the transformed cancelable iris template through the proposed method. In a nutshell, three methods have been proposed in this thesis to improve the authentication performance and flexibility of protected iris recognition systems.

Actions (login required)