A robust preprocessing algorithm for iris segmentation from low contrast eye images
Joke A. Badejo, Aderemi A. A. Atayero, T. S Ibiyemi · 2016
Iris recognition systems offer highly accurate personal identification both on small and very-large scale systems needed in government, forensic and commercial applications. The automatic segmentation of a noise-free iris region is imperative for optimal performance of the system. However, image characteristics such as brightness and contrast, the differing levels of pigmentation, occlusion by eyelashes and/or eyelids, coupled with varying sensor and environmental conditions, makes iris segmentation a huge and difficult task. This paper proposes an image pre-processing algorithm for robust iris segmentation of low contrast images, aimed at reducing mis-localization errors of basic curve-fitting algorithms. Similar to face detection, the algorithm performs iris detection with a k-NN classifier trained with features extracted by a rotation-invariant texture descriptor based on the co-occurrence of local binary patterns. The integration of the proposed algorithm into an existing open-source iris segmentation module offered a 40% improvement in execution time; a segmentation accuracy of 92% was also recorded over 1,898 low contrast eye images acquired from African subjects. The low contrast eye images were acquired to support diversity in iris recognition.