Research On Iris Identity Recognition Algorithm

In the informatization time, how to correctly identify a person’s identity and protect his information safe already becomes a social problem which needs to be solved. The traditional identity autentication methods become more and more difficult to satisfy people’s requirements as a result of its easy forging and lost. At present, the most convenient and fast solution will undoubtedly be the biometrics. Because iris recognition has the least false recognition rate among all kinds of biometrics, and it has the merits of uniqueness, stability, non-invasive and antifalsification, iris recognition has the broard application prospect, huge economic benefits and research value.Iris recognition algorithm includes iris image quality assessment, iris image pre-process, feature extraction and pattern matching, while iris image pre-process includes the localization of inner and outer circles, eyelid localization, iris normolization, image enhancement and noise removal.In this paper, we mainly research and innovate following iris recognition parts:1) Usually there are defocus and eyelid occlusion in the obtained iris images, which increase difference among intra-class iris images, and at last will make the false reject rate of recognition system up. To reduce the complexity of algorithm and computation, we propose a fast and efficient multiple step iris image quality assessment algorithm, which is as: Firstly the high frequency power of iris local areas lying in the two sides of the pupil is calculated by Laplacian of Gaussian operator to discard the defocused images, secondly the average gray value of the upper designated area of the pupil is computed to discard the seriously occluded images by eyelid, lastly the occluded degree by eyelash is measured through the horizontal high frequency power of the upper designated area of the pupil. Besides, the algorithm only needs to localize pupil, avoids the hard localized and time-consuming outer circle localization. In addition, the algorithm just extracts relative information from iris local areas to assess image quality, needs not to handle the whole image. We experiment it on CASIA 1.0 and CASIA 2.0 ver1 iris libraries, and the result shows that it is a fast and efficient iris image quality assessment algorithm.2) In localizing iris, we propose a fast, accurate and robust iris localization method, while the classical methods have disadvantage of large calculated amount and consuming time because of their iteration in three-dimension parameter space. According to the gray distribution of iris image, the gray of sclera, iris and pupil is of ladder distribution, and it is obvious between pupil boundary and iris boundary, so inner circle of iris is easy to localize, and we separatsely use the least squares method and geometric method to localize pupil. Then on the basis of pupil localization, we use the improved Canny operator plus Hough transform to localize outer circle of iris. While detecting eyelids, we propose a method of Radon transform subsection line to localize eyelid, and use threshold method to remove eyelash and eyelid shadow.3) As a result of different sizes and different resolutions of iris images and translation distortion, zoom distortion and rorate distortion with different degrees in iris images, it is hard to compare directly iris patterns. We use Daugman’s Rubber Sheet Model to normalize the licalized iris, and transform the annular iris into the same size rectange which is of size 512×64. To remove the eyelid and eyelash interferences, we select the right corner quarter of normalized iris as the valid area to do the subsequent recognition, to avoid influence from pupil, we select iris pixels from the fifth row, and the valid iris area is of size 256×32.4) Because 2D Gabor function could simulate the feeling characteristics of a pair of simple visual neurons from animals, we can use it to extract useful pattern feature through selecting different window, frequency and direction parameters. In this paper, we design and construct 32 enterclose Gabor filter based on polar form to extract pattern feature from normalized iris image, and then we use Hamming distance to match. We select 53 eyes, 371 images from CASIA 1.0, whose useful iris area is disturbed a little, and when threshold is 0.372, the false reject time is 18, the false acception time is 587, the false rejection rate is 1.62 %, the false accept rate is 0.87 %, and the correct recognition rate is 99.119 %, while the result is pleasant.