Advancing the technology of sclera recognition

Abstract

Emerging biometric traits have been suggested recently to overcome some challenges and issues related to utilising traditional human biometric traits such as the face, iris, and fingerprint. In particu- lar, iris recognition has achieved high accuracy rates under Near- InfraRed (NIR) spectrum and it is employed in many applications for security and identification purposes. However, as modern imaging devices operate in the visible spectrum capturing colour images, iris recognition has faced challenges when applied to coloured images especially with eye images which have a dark pigmentation. Other issues with iris recognition under NIR spectrum are the constraints on the capturing process resulting in failure-to-enrol, and degradation in system accuracy and performance. As a result, the research commu- nity investigated using other traits to support the iris biometric in the visible spectrum such as the sclera. The sclera which is commonly known as the white part of the eye includes a complex network of blood vessels and veins surrounding the eye. The vascular pattern within the sclera has different formations and layers providing powerful features for human identification. In addition, these blood vessels can be acquired in the visible spectrum and thus can be applied using ubiquitous camera-based devices. As a consequence, recent research has focused on developing sclera recog- nition. However, sclera recognition as any biometric system has issues and challenges which need to be addressed. These issues are mainly related to sclera segmentation, blood vessel enhancement, feature ex- traction, template registration, matching and decision methods. In addition, employing the sclera biometric in the wild where relaxed imaging constraints are utilised has introduced more challenges such as illumination variation, specular reflections, non-cooperative user capturing, sclera blocked region due to glasses and eyelashes, variation in capturing distance, multiple gaze directions, and eye rotation. The aim of this thesis is to address such sclera biometric challenges and highlight the potential of this trait. This also might inspire further research on tackling sclera recognition system issues. To overcome the vii above-mentioned issues and challenges, three major contributions are made which can be summarised as 1) designing an efficient sclera recognition system under constrained imaging conditions which in- clude new sclera segmentation, blood vessel enhancement, vascular binary network mapping and feature extraction, and template registra- tion techniques; 2) introducing a novel sclera recognition system under relaxed imaging constraints which exploits novel sclera segmentation, sclera template rotation alignment and distance scaling methods, and complex sclera features; 3) presenting solutions to tackle issues related to applying sclera recognition in a real-time application such as eye localisation, eye corner and gaze detection, together with a novel image quality metric. The evaluation of the proposed contributions is achieved using five databases having different properties representing various challenges and issues. These databases are the UBIRIS.v1, UBIRIS.v2, UTIRIS, MICHE, and an in-house database. The results in terms of segmen- tation accuracy, Equal Error Rate (EER), and processing time show significant improvement in the proposed systems compared to state- of-the-art methods.