Non-cooperative Iris Recognition Research,

In information era, with the growing needs of security, people take more and more concerning about biology identify technologies which use a person's biometrics characteristics to identify. In all these biology recognition technologies, iris recognition technology has the prime advantage. As an important identification characteristic, it maintain stable over life and variable between persons. No body can have an iris the same with the other person and even if the two irises are from the same person's left and right eyes are different. Iris has random texture, which grown in one's childhood and last the same in entire life. It has high stability. Add to the fact that no need to touch with the recognition equipment, Iris recognition is an ideal non-invasive recognition technology.In this thesis, the development of Iris recognition technology and applying prospect from home and abroad was introduced, and the basic principle of iris recognition was illustrated. Based on analyze of the development trend of iris recognition technology, non-cooperative iris recognition technology has been researched. The key point of the research is the improvement of the algorithms, which made it robust to illumination, eyelid or eyelashes covered and reflections.The main work was accomplished in the paper as following: In the step of iris locate, based on the characteristics of the noisy image, adopted a method which firstly locate the outer boundary of iris and secondly locate the inner boundary of iris. The algorithm for locating is an improvement algorithm based on Hough transformation. The Canny edge detector was improved by add a parameter to indicate the direction of the edge points supposed to distribute. The parameters space of Hough transform was reduced by shrink the iris center range. The occluded eyelid was located using Radon transformation. In the normalized step, using a dynamically threshold way to eliminate the reflection and eyelashes. In the characteristic extract step, adopted a multi-scale filter based 2-D Log-Gabor wavelets. In the match step, a hamming distance classifier was constructed. The algorithm has been tested in MATLAB environment using the iris image from UBIRIS. The experiment result shows the algorithm proposed in this thesis is effective and has good performance.