A novel approach for eye gaze and tilt estimation

Abstract

Conventional iris biometric system, in its localization module, detects iris boundary through integrodifferential intensity change among concentric circles drawn on pupil center. The pupil is detected as the darkest region in human eye. However, this crude approach performs well for constrained iris images captured in near infrared (NIR) spectrum, but may fail for low-quality color eye images captured in visible spectrum (VS). This paper proposes a novel approach that estimates eye gaze and tilt without precise knowledge of pupil location. Rather the proposed technique color-segments the sclera region to find few low-cost nodal points within eye region. The proposed method localizes sclera through a novel color segmentation method applied in YCbCr color space. In the next phase, during content retrieval process of the sclera, typically six nodal points are extracted whose relative positions define the gaze and tilt of the eye of the subject. The Proposed method has been experimented on 100 randomly chosen images from UBIRISv2 unconstrained VS iris database. The experiment yielded 96% accuracy in proper sclera-localization and extracting the six low-cost nodal points.