Iris Biometric Extraction And Recognition

An effective identification for true identity of person is need in the field of national security, information security, financial security, aviation safety and so on. However, biometric recog-nition technology is the most effective way to solve this problem. Iris has rich and unique texture feature, which is an idea approach for personal identification. Due to biometric recog-nition based iris has better reliability, high recognition rate, natural security, and has wide ap-plication prospect. Meantime, it has been received more and more attention by the academic circles and the business community, which becomes a very active research area in the biomet-ric recognition. In this thesis, the key technologies of iris recognition are studied from the physical structural characteristics of iris itself. At the same time, some new idea and algo-rithms of iris recognition are proposed. The main work and contributions of this thesis are as follows:1) Iris image quality assessment is studied in the thesis. Considering the input of poor iris images will affect the recognition performance, image quality need to be assessed. Only im-ages that meet certain quality can be used to identify. On the basis of the existing methods, an objective iris image quality assessment algorithm is proposed by analyzing the factors affect-ing the quality of iris image and the specific characteristics of iris image. The method evalu-ates the clarity of image from overall clarity and texture one, and calculates the effective area of iris from the visibility of crude assessment and fine assessment. It not only can quickly recognize the badly quality image but also evaluates an image from quantitatively.2) The preprocessing of iris image is studied in this thesis. The existing iris preprocess-ing algorithms are of low executing speed, poor robustness and accuracy, an improved pre-processing approach is proposed at the basis of the predecessors'studies. Iris image process-ing includes iris location, normalization, denoising and enhancement. Especial, iris location is an emphasis in the preprocessing, it includes inner and outer boundary of iris location, eyelid and upper eyelash detection. A location method of coarse-to-fine is presented on the basis of the existing models and algorithms, then, the search domain is reduced and the location speed is improved. Edge detection combing with least squares to locate rough pupil, using active contour based without re-initialization level set method to locate its fine boundary. The outer boundary is located with the improved Hough transform method on the basis of rough pupil location. The upper and lower eyelids are detected by fitting a parabola with least-squares. Most upper eyelashes can be quickly detected with double threshold method. The preprocess algorithm tests on four kinds of different iris databases and evaluate the algorithm perform-ance with the location time and accuracy rate.3) Iris feature extraction and matching methods are studied in the thesis. Extracting stable and distinguish iris features is the key of iris recognition. Three different iris feature extrac-tion and recognition methods are proposed in this thesis. The proposed algorithms have po-tential advantage, it compared with the traditional iris recognition methods, Firstly, in view of curvelet transform decomposes image from different scales and direc-tions, which adequately describe the image texture feature of a straight line or curve. So, an iris feature extraction method based curvelet transformation is put forward. Different feature extraction ways are adapted according to information that the low frequency and high fre-quency sub-bands presented. The algorithm performance is verified in the verification mode and identification mode. Compared with traditional transform-based extraction methods, iris features extract from curvelet coefficients can be better presented the mainly iris features. Secondly, due to contourlet inherited the anisotropic multi-scale relations of curvelet trans-form, which can extract the important intrinsic geometry features of image. Iris texture feature extraction method based on contourlet transform is proposed. The method uses three different structure to extract iris texture features that the mean, variance, moment invariants and energy from different scales and directions of sub-band coefficients. Support vector machine classi-fier and distance-based matching method are adopted to recognition. The experimental results demonstrate that the iris texture feature information can be effectively extracted by the pro-posed method.Finally, considering of empirical mode decomposition (EMD) has a multi-scale character-istic, which can adaptively decompose the signal and effectively extract the global and part information. At the same time, EMD can overcome the difficult of selecting wavelet basis function when wavelet transform is used to extract feature. So, an iris feature extraction method based on EMD and singular value decomposition (SVD) is presented in this thesis. The decomposition features using SVD to decompose element of intrinsic mode functions, which can effectively describe iris feature and reduce the feature dimensions. Meanwhile, compared to the feature extraction of wavelet based Garbor and Harr, code time and length are clearly improveed. At last, Modest AdaBoost classifier is adapted to four kinds of iris im-age database. The recognition rate reaches 98.9% when it tested on the iris image database of CASIA Ver1.0. Results show that iris features extracting by EMD have better ability to dis-tinguish other iris.The results show that feature extracted by EMD have good discrimination, which can obtain satisfied identification result.