Iris Recognition For Practical Environment

Information security technology is paid more and more attention in information society. There are some defects of the traditional identification, such as fragile, easily lost, easily stolen and easily cracked, which has been unable to meet the needs of modern information security. Biometrics identification technology which is based on the human physical characteristics or behavioral traits for identification makes up the deficiencies of traditional identification methods. Iris recognition has more obvious advantages in the exclusiveness, stability, universality, accessibility and acceptability than other biometric identification technology. So this article will take iris recognition as a researching object.This article examines the validity of iris recognition algorithm which is based on a practical environment; it also discusses and researches how to improve the recognition effect which is used to describe the iris recognition technology in practical environment. Firstly, this article introduces the principle and research trend of iris recognition system; Secondly, it elaborates iris collection, preprocessing, feature extraction, matching methods and implementation steps according to several processes of iris recognition in details; thirdly, it will extract some iris samples to simulate practical environment from the JLU-IRIS iris images library which are collected by us. We can get some recognition effect from experiments. The results reflect that the methods which are referred to above are interfered obviously. To solve this problem, this article introduces a new method which is called the topology-based pattern recognition of iris recognition; also we give a new way to carry out this method. The experiment proves that this method can improve the robustness of iris recognition in practical environment, and it achieves the desired aim. The main word and achievement of this article is as following:1. Using the iris image acquisition instrument of independent research lab to build JLU-IRIS iris image library, JLU-IRIS iris library collects the 271 collected persons from the iris image sequence of right and left eyes under different time intervals, different illumination environments as well as different angles in long time gathering cycle. (Iris image video).When each iris image is collected, we adjust the indoor incandescent light to adapt the environment luminous intensity. At the same time, the collected persons carry out small-angle offsetting faces and eyes intentionally for simulating iris translation and revolving situation. Collections for each image sequence lasts more than 5 seconds, per second video stream is combined with the 25 frames. We have adopted the image of the video stream interception frame operation, and it has obtained more than 203.250 iris images. Finally, we unify the storing format of the iris images and form the JLU-IRIS iris database.2. The pretreatment of iris images is achieved. The pretreatment of iris image is divided into three parts: iris localization, normalization and image enhancement. We localized iris from inner to outer. First of all, According to the coordinates of troughs besides the maximum peak in the row and column's histogram of iris image, this article parts the probably scope of iris's inner from the whole image. Then it searches the edge points in the scope of iris's inner by Canny operator. This algorithm uses first derivative smooth image of Gaussian function combining the Gradient algorithm which can detect the specific direction's gray point mutation to discriminate the edge point. In the following, Hough based on policy of voting is used to detect the inner pixels of iris from the possible peripheral points which are acquired by Canny, and at the same time the parameters about the inner circle of iris are obtained. In the process of the outer border detection, according to the results of the inner location, in the direction of the line passing the center and the fan-shaped area of iris, calculates the positions where the Max of the adjacent pixels' gradient are located, and then use these locations to get the parameters about the outer circle of iris. In normalization, it converts the ring-shaped iris area to fixed size rectangle area which is based on the affine operation of polar coordinates to rectangular coordinates. The values of pixels whose coordinates are not integer should be transferred from the pixels of the former four pointers by bilinear interpolation. In the end, it deals with the influence of ambient light by the means of linear extrusion in gray level and enhancement of partial iris image, and strengthens the texture of iris.3. Three traditional methods of feature extraction and feature-match are realized, and the Iris images of JLU_IRIS database not only in ideal ,but also in true environment are used as samples to test the effect of different algorithms. The three traditional methods are composed as follows: the Gabor use the method of Fourier in multi-scale and multi-direction to extract the phase info of iris images, and the filtering results are used as the eigenvector, and Euclidean distance is used to judge different classes. The zero-crossing location are detected after the multi-scale wavelet transform, and the eigenvector are composed by zero-crossing locations and scale correlation, and the dissimilarity function is used to judge different classes of Iris image. The PCA/ICA model is selected to train the subspace of features according to the classified iris images. In the process of recognition, the sample is firstly mapped to the subspace of features, and then use the method of cosine distance to classify the sample .The results show that: in ideal condition, the mistake rate of Gabor is 2.98%, and that of zero-crossing position of wavelet transform is 5.51%,and that of PCA/ICA is 1.37%.But in the true environment ,the mistake rate of Gabor is 14.44%,and that of zero-crossing position of wavelet transform is 23.51% ,and that of PCA/ICA is 22.17%. The results show that the many interference of true environment make robustness lower than in the real environment. Of all the algorithms, the Gabor is not so sensitive.4. For a variety of effects in practical environment caused decline in iris recognition, we introduce the topology-based pattern recognition method to iris recognition. First we introduce the theoretical basis for topological pattern recognition, and relevant definitions in high-dimensional space. Then we introduce the same samples'continuity in high-dimensional space which is the starting point of this method. We propose implementations which base on sausage body covering with samples of known structure topology, and how to determine the process of iris recognition achieving through this topology. The experiments prove that the method in practical circumstances, if the sufficiency of training samples can be offered, the better results with FAR is 0, and 4% to FRR, 96% to CRR.In summary, the processes of the collection of iris recognition, pretreatment, feature extraction and matching are researched in this article. At the same time, we describe and analyze the algorithm in these steps. Through the experimental results, we prove that a larger gap between practical environment and ideal environment in iris recognition effect, the interference of environment impact the robustness of iris recognition. For this phenomenon, we introduce topological pattern recognition into iris recognition. And propose a method to achieve this idea, and authenticated the method by experiment. The practical environment can also achieve good effects of iris recognition with this method. It offers foundation of future practical application in iris recognition.