Abstract
Pupil localization in human face/eye images has numerous applications, e.g., eye tracking, iris recognition, cataract assessment and surgery, diabetic retinopathy screening, neuropsychiatric disorders diagnosing, and aliveness detection. In real scenario, the pupil localization task suffers from many complications such as pupil's constriction and dilation moments, light reflections, eyelids and eyelashes, and cataract disease. To resolve this issue, this study proposes an accurate and fast pupil localization scheme. It performs relatively well for eyeimages acquired either with the near infrared (NIR) or visible wavelength (VW) illumination. First, it effectively preprocesses the input eyeimage. Next, it coarsely marks pupil location using a scheme comprising an adaptive threshold and two-dimensional (2D) object properties. Then, it validates pupil location via an effective test involving global gray-level statistics. If it finds pupil location invalid, then it localizes pupil through a hybrid of the Hough transform and image global gray-level statistics. Finally, it localizes the fine pupillary boundary through a hybrid of the Fourier series and image's gradients. Its experimental results obtained on numerous publically available iris datasets demonstrate its superiority over most of the contemporary schemes.