Combining multiple iris texture features for unconstrained recognition in visible wavelengths

Abstract

Recognizing individuals in unconstrained environments without their cooperation, e.g. based on surveillance imagery, represents an active field of research. Non-cooperative iris recognition based on images captured at visible wavelength (VW) represents an extremely challenging task. To enhance the reliability of recognition systems using VW iris images, and to enable their use in forensic applications, state-of-the-art approaches additionally employ periocular information surrounding the eye. Hence, a fusion of information obtained from multiple biometric characteristics is vital in scenarios where the quality of biometric samples is severely affected by numerous factors. In this paper, we investigate the potential of a multi-algorithm fusion employing iris texture information extracted from VW iris images. Features extracted by four types of algorithms, i.e. conventional methods, keypoint-based methods, generic texture descriptors, and colour-based methods, are combined to improve the recognition accuracy. By performing a weighted score-level fusion of comparison scores obtained by four different types of feature extractors, improvements in biometric performance of 15% and 27% are achieved on subsets of the publicly available UBIRISv2 and MobBIO iris database, respectively.