The Research Of Iris Segmentation Algorithm

Iris segmentation plays the most important role in an iris biometric system and its accuracy will impact the comprehensive performance of the system to a large extent. At present, the algorithms described in opening literatures about iris segmentation all may have more or less defects. An excellent iris segmentation algorithm should be able to handle strong noises and have a high speed and accuracy in processing. In the iris image acquisition phase due to the complex environment the image may have a low quality, such as, non-uniform lighting, motion blur, iris off-axis, the iris occluded by eyelids, etc., all these interference factors increase the difficulty of complete iris segmentation and also put forward the higher demands to the performance of the iris segmentation algorithm.A complete iris segmentation algorithm contains the specular reflections removal, the localization of pupillary and limbic boundaries, the eyelid boundaries positioning and eyelashes elimination, etc. This paper focuses on the iris segmentation algorithm and improves the existing methods and proposes new model to boost the accuracy of the iris segmentation. At the same time, the iris recognition system is also constructed based on the segmentation algorithm proposed in this paper to verify the effectiveness of the proposed algorithm. The main work of this paper is summarized as follows:Propose an average linear interpolating method to remove the specular reflections. This method uses the information provided by the pixel grayscales around the reflection to calculate the value of interpolating point. The side which the interpolating point more close to it will provide more information for interpolating while the further provide less. This is a more reasonable method compared with the other simple inpainting methods.Combine the Haar-like features with the Adaboost cascade detector to extract the ROI(region of interest) from original iris image. In order to evaluate the performance of the constructed classifier, the CASIA-Iris V3-Lamp and the LZU-iris iris databases are used for testing the classifier and the precision rates are 95.57% and 98.26%, respectively.Propose a rough pupil center positioning method using contour detection and local adjustment. In processing, a binarization method based on the accumulation of the histogram of the ROI image is put forward to separate out the pupil region. The local adjustment can effectively overcome the locating offset caused by gray centroid migration to some extent.Propose a novel pupil edge points detecting and screening method and use the least square method to approximate an elliptical shape as the edge of the pupillary boundary. Adopt the improved circular Hough transform to locate the limbic boundary. Here, through shrinking the parameters space of the searching circle to increase the efficiency of the algorithm. At last, The Canny edge detection method and least square method are used for eyelid segmentation.Construct the iris recognition system based on the iris segmentation algorithm proposed in this paper and the CASIA-Iris V3-Lamp iris image database. Through evaluating the performance indicators of the system(such as, FAR, FRR, ERR, ROC) and comparing the experimental results with the opening research literatures to prove the effectiveness of the proposed algorithm. The ERR of the iris recognition system in this paper is 3.25 %.