A low complexity Iris localization algorithm for Iris biometrics

Abstract

Though there has been a plethora of the iris localization schemes, most of these have not been implemented in the real-time systems due to computational complexity. It is mainly because researchers often use the Integro-differential operator (IDO), circular Hough transform (CHT), active control models (ACM), and/or machine-learning (ML) techniques to mark iris in the human eyeimages. While these schemes exhibit relatively better performance, most of these generally take longer due to complex architecture. To contribute to this concern, authors propose a low-complexity iris localization algorithm that works as follows. First, it suppresses the light/specular reflections and sharp gray level variations in the input eyeimage using an order statistic-filter. Using a coarse-to-fine scheme, it locates potential edges in the gradient eyeimage. Next, corresponding to each edge, the gray-level intensity of a circular region is examined. If a compact region having lowest gray-level intensity is found, then it is declared as pupil and its non-circular boundary is extracted using the 8-connectivity method. Finally, using a coarse-to-fine scheme, it marks two safe-regions in input eyeimage, draws a set of three parallel white radial-segments in each safe-region, gets gradient image via the Canny edge-detector, detects potential points on the edge of limbic (iris outer) boundary using the concept of intersection-point of the radial-segment and limbic boundary. The proposed scheme is validated on the MMU V1.0, MMU 2.0, IITD V1.0, CASIA V1.0, CASIA-IrisV3-Twins, CASIA-IrisV3-Interval, CASIA-IrisV3-Lamp, and UBIRIS V1.0. While exhibiting tolerance to noisy regions (e.g., eyelids), it takes less than a second to mark both iris contours, which is a green signal for its real-time applications.