Enhanced segmentation and complex-sclera features for human recognition with unconstrained visible-wavelength imaging

Sclera recognition has received attention recently due to the distinctive features extracted from blood vessels within the sclera. However, uncontrolled human pose, multiple iris gaze directions, different eye image capturing distance and variation in lighting conditions lead to many challenges in sclera recognition. Therefore, we propose an enhanced system for sclera recognition with visible-wavelength eye images captured in unconstrained conditions. The proposed segmentation algorithm fuses multiple color space skin classifiers to overcome the noise factors introduced through acquiring sclera images such as motion, blur, gaze and rotation. We also propose a blood vessel enhancement and feature extraction method which we denote as complex-sclera features to increase the adaptability to noisy blood vessel deformations. The proposed system is evaluated using UBIRIS.v1, UBIRIS.v2 and UTIRIS databases and the results are promising in terms of accuracy and suitability in real-time applications due to low processing times.