| With the development of information technology in contemporary society, the demands for security are becoming more and more important. So the technology of biometrics identification was improved and applied on many fields. Iris Identification is one of potential biometrics technology; it has been gradually applied to safety fields. Iris Identification is based on iris texture to identify persons, and it is one of most accurate biometrics technologies.Based on the theory and method of image processing, signal analysis and pattern recognition, several parts of iris recognition are explored systemically in the paper. First of all, a standardized sample database of iris images that named as JLU-IRIS 1.0 version was set up, it is as one of the experimental samples of iris recognition algorithm in the paper, and then in the regions of iris localization, iris preprocessing, and feature extraction and matching, we all propose new algorithms. Finally, in order to facilitate testing various iris recognition algorithms, the object oriented layer class structure unify exoteric iris recognition system is proposed and designed. And the iris recognition system is developed on the system platform under the framework. All the algorithms which are proposed in the paper are implemented and tested. The experimental results show that the proposed algorithms achieve encouraging recognition performances.In this paper the following issues have been identified and explored:1. Standardization of iris image database construction. First of all, we collect 271 different individuals of right and left eyes iris images videos in different times (morning, afternoon, evening, each time interval of about two weeks) using the iris acquisition system which is developed and set up by ourselves and convert the 1,626 iris videos to 225,750 pieces of 8 bit 256 grayscales bitmap. And then we evaluate and select each class of iris images using sequence image quality evaluation algorithm which is proposed by our research group. Lastly, through the unified numbering for iris images and storaging, a standardization of iris image database named as JLU-IRIS 1.0 version which has a total of 2,710 pieces of iris image of 271 classes is established. We will share the iris image database to iris recognition algorithm researchers at home and abroad. 2. In the iris location respect, this dissertation gives one new iris location algorithm based on active contour model which named as Snakes. First the binarization iris image is narrowed search region using the ranks of the mean gray-scale projection, and gets the smallest region image that contains the pupil. The image is transferred into binary image by dynamic threshold; the threshold can be obtained by analyzing of gray histogram. So, the pupil can be found. The boundary data are described using Fourier Description (FD). It is as the initial data set of the Snakes, and then carries out precision positioning of the pupil using Snakes. At last, the search region of iris outer is determined in accordance with the pupil, the iris outer is positioned by differential cumulative gradient algorithm. By using this algorithm, the results show that the time cost in the iris location is sharply shorted, and the precision is improved a lot.3. In the paper, a new iris preprocessing algorithm is proposed based on rough matching of corners. First of all, Corners of iris texture are detected using Harris method, and then The relationship set of corner matching is established between template image and identifying image using the cross-correlation coefficient algorithm. The rotation offset value is calculated between corresponding corners that have different angles with horizon, the average of rotation offset value is the matching result between template image and identifying image. The experimental results show that the time cost is sharply shorted; the proposed algorithm can effectively reduce the matching scale for further recognition. The matching result can also be as the initial rotation angle in normalized and eliminate the influence of rotation.4. In the iris normalization and enhancement respect, Iris normalization is based on a new algorithm that combines the center of pupil and iris to one point. The normalization algorithm can correct the error caused by the reason that centers of inner and outer circles are not same. Considering the practical characteristic of iris image, improved local histogram equalization is used for image enhancement. This method can reduce the influence of uneven illumination.5. Aiming at the shortcoming that single iris feature recognition method can not express the iris feature comprehensive; the dissertation introduces the data fusion technology, the framework of iris recognition is proposed based on pretreatment matching of corner. It requires corner matching firstly, and then identifies using other feature methods according to judgment. The dissertation further presents the iris recognition fusion algorithm based on linear fusion method. Several feature recognition methods are selected to classify. These matching results are fused using linear fusion method. The dissertation mainly introduces the Gabor filter iris recognition method and Wavelet zero-crossing iris recognition method, the experimentations and results show that the proposed algorithm can shorten the recognition time and decrease the FAR in the framework, the fusion algorithm can overcome disadvantage of single algorithm ,improve the robustness of iris recognition and increase the recognition rate.6. In order to facilitate the different methods of iris recognition and processing algorithms can be conveniently sharing, reuse and extension. The object-oriented open-tiered framework of iris recognition system is proposed in the paper. The whole process of the iris recognition is divided into several functional modules according to the layer of the iris recognition, these modules communicate with each other through standard data interface. Other modules are not changed if algorithm of one module is changed; the framework is a typical high cohesion, low coupling model. The proposed algorithms in the thesis are implemented based on the platform of the framework, and then a large number of experimental tests are carried out, the results show that the proposed algorithms in the paper are effective.To sum up, I have made a lot of work and in-depth study in standardization of the iris image database building, iris location, the improvement of iris preprocessing, feature extraction and matching and designing framework of iris recognition system. And I have designed a prototype iris recognition system; the algorithms and system have been evaluated extensively on JLU-IRIS 1.0, Experimental results shows the system performs very well. |
PDF Full Text Request