The Research Of Iris Recognition Based On Fractal Multi-resolution Blanket Dimension Geometry

Nowadays biometric technology attracts more and more attentions of different countries and many fields with global safety and practicality being highlighted. As a kind of biological characteristics, iris recognition has become the focus of experts and scholars with its advantages of high accuracy and non-invasive. In the last twenty and thirty years, iris recognition technology develops rapidly and becomes mature gradually. It has been considered to be the most reliable and promising biometric technology.This paper introduces the development state and categories of biometric recognition technology firstly. Then the components of iris recognition system are described, and some classic iris identification algorithms are summarized. This paper focuses on image preprocessing, feature extraction and pattern matching.The iris area is located based on gray characteristics of iris. First, a point within the pupil is found according to the lower pupil gray value. Because the gray curvy presents the highest and lowest points at the inner and outer boundary, we search three points on the edge of a circle which are not on the same line. At last the inner and outer boundary can be located according to the principle that three points which are not on the same line can define a circle. The locating method can achieve a higher location speed with lower error.The features of blanket dimension, blanket dimension intercept and multi-resolution blanket dimension are extracted for iris recognition, and tensor subspace analysis is also used to characterize iris texture. The blanket covering technology can reflect the change of iris texture. Multi-resolution combines the resolutions changing from high to low into a vector which can further describe texture changes at different resolution. And blanket dimension intercept can reflect the change speed of image gray level and rough degree of image surface. The manifold learning methods based on tensor subspace analysis can effectively detect the intrinsic geometrical structure of iris texture. Experiments show that multi-resolution blanket dimension can achieve the most effective recognition performance.Normalized correlation classifier is used for pattern match. And a rotation compensation algorithm based on1D rank filter combined with parabolic fitting using least squares is introduced to solve the problem of iris rotation. Compared with traditional methods of cyclic shift, a rotation compensation algorithm proposed in this paper can effectively improve the performance of iris system.At last we make the conclusion of this paper. All the above algorithms are simulated on the platform of Matlab7.0with CASIA-IrisV3-Interval iris database. The experimental results show that the proposed algorithms can reach a good performance.