Iris Localization Algorithms

Iris recognition is a hot topic in applied mathematics, image processing and pattern recognition, since it has high reliability and robustness. Iris localization is a method that segment iris region from the capture images. Accurate iris localization is a prerequisite of the efficient iris recognition system. In this paper, the main content is how to improve accuracy and robustness of localization algorithm, while reducing the time required.In this paper, the structure and working principle of iris recognition system are introduced firstly, and then four important existing iris localization methods are elaborated. Furthermore, main problems about existing localization algorithms are analyzed.For the existing problems, we propose an effective iris localization algorithm which consists of the following process:pupil's location estimation, noise detection, the iris inner boundary localization and the iris outer boundary localization. Iris inner boundary localization uses a coarse-to-fine process, while in the outer localization the radius of outer circle is estimated firstly, after which fine localization is performed. Localization is executed in the outspread image in polar coordinate, the method that circular boundary detection in the original image is transformed into detecting line boundary in the outspread image, improves the efficiency of the algorithm. Considering the difference of intensity value in local area when detects boundary in outspread image, we detect the boundary by weighted integrodifferential operator which the gradient is obtained by the weighted average of intensity value in N adjacent rows. Using the weighted integrodifferential operator not only reduces the influence of noise point, but also strengthens the boundary information of the iris. Then the accuracy and robustness of iris localization are improved.The experiments on CASIA-V3.0 show that the proposed algorithm has a higher accuracy and robustness. The algorithm has also made a satisfactory performance on poor quality iris images, overcomes the shortcomings of existing algorithms which require ideal iris images. Moreover, it has a higher computational efficiency.