Novel technique for removing corneal reflection in noisy environment — Enhancing iris recognition performance

摘要

The pupil localization is inaccurate due to noisy artifacts, resulted by corneal reflection present in an iris image. Hence, the performance of iris-based recognition system is degraded. This paper describes a unique approach to detect noisy pixels that are present in the pupil. This technique is based on thresholding. The detected reflection noisy pixels are replaced with the neighborhood non-reflection pixels. Finally, local binary pattern (LBP) is used for extracting texture features and they are matched using radial basis function and probabilistic neural network classifiers. To evaluate the performance of the proposed technique, the approach has been applied on PHOENIX, MMU, IITD and CASIA 4.0 databases. The system accuracy of 97% is obtained by our integrated approach of removing corneal reflection algorithm and probabilistic neural network classifier for PHOENIX database.