Research And Exploration Of The Iris Recognition System

With the rapid development of digital and information society, advanced requirements on security has been proposed by nations and individuals. However, traditional methods for personal identification include the token based methods that use specific things such as ID card or keys for authentication and the knowledge based methods that use something you know, passwords, for instance for identification are not so reliable as you thought. A new method for personal identification named as biometrics is born and has attracted more concerns recently. Iris with the characteristics of richness, uniqueness, stability, noninvasiveness, and key point high confidence has great potential.The iris recognition system includes four parts: image acquisition, iris image preprocessing, feature extraction and matching. Image acquisition- design a device to capture the original iris images; image preprocessing- include the localization of the pupil (inner) and the limbus(outer) boundaries of the iris, getting rid of noises, eyelid, eyelash, strong reflections for example, normalization and enhancement; feature extraction- get feature vectors to represent the iris images from its abundant texture information and code; matching- classify the feature vectors through different algorithms like Hamming Distance, weight vector, etc and then match with the irises in the database.This dissertation got a few fruits through the research of abundant related literature and plentiful experiments. Its primary contributions are as follows:1. Image quality evaluation methods were investigated in detail and executed in iris image definition, noises of eyelash and eyelid, and movement of pupil's aspects.2. In order to decrease computational cost, a geometrical method was used to calculate the pupil's radius and center coordinates through choosing three points from the bottom part of the circle, for there was little noise. The iris image was shrinked by a certain rate (30% in this paper), Canny operator in vertical orientation was used to detect the edge, the top eyelid region was cut and pupil area was excluded, only the left and right 45 degrees'arc was remained, and the radius and center coordinates of outer circle was calculated using improved Hough transform.3. Noises like eyelid, eyelash, strong reflections etc were processed, which reduced the iris image matching error rate.4. A novel algorithm for coarse classification of iris images using a box-counting to estimate fractal dimensions method combined with improved BP neural networks was proposed, and experiments showed that our method saved so much of iris searching time. Because of the restriction of bandwidth of Gabor wavelet, this dissertation used 2D log Gabor filters to extract iris features.All the algorithms mentioned above have been emulated on MATLAB6.5, the experiments show encouraging results that the matching speed has been improved greatly. The result of the research can be used in airdrome, bank etc, and it will make deep influence on these fields.