Research Of Iris Recognition Algorithm And Its Realization

With the quick development of intemet and information people require more demanding information,so people need more correct,safe and practical technique.The traditional security technology does not meet the requirement of security quality of the people at present. Then the people turn to biology recognition technology and make it into a fast developing phase.Iris recognition technology with its huge potent is relatively new among different kinds of biometric recognition characteristic of an iris all through one's life,and it can't forge and infringe.In this paper,the development of biometric recognition and some kind of representative biometric techniques are introduced.Then the development and meaning of iris recognition technique and the structure of iris recognition are discussed in detail.The algorithm of iris recognition has also been researched deeply and of the processed including 4 steps:location, normalization,feature extraction and pattern mach.My thesis focuses on normalization and match.In the process of normalization,a method based on Daugman's plastic sheet model is applied,resulting in transforming a circle area to a rectangle area.This method made several segments representing two different area between circles,as long as sampling points on segments we can get good normalized effect,using only one analysis model.It reduces blindness in searching,saves time,and decreases complexity in localization.Hough transform is employed to exact localizationIn the process of encoding and matching,firstly putting 2 dimension iris image to 1 dimension grey scale signal,reducing calculation;secondly,using 1D Log-Gabor filters extracting binary iris character vector,reducing complexity of algorithm and increasing recognition speed;fmally improved Hamming distances are employed to match,resulting in rotation invariable.This thesis uses the Chinese Academy of Science-Institute of Automation eye image database(CASIA 1.0),choosing 80 people times 7 images=560 images.Though a great lot experiments we have proved this algorithm:in CASIA-a the centre wavelength of 18 pixels is optimal for the dataset.An optimal template size with radial resolution of 20 pixels,and angular resolution of 240 pixels was chosen for the data sets.These parameters generate a biometric temple that contains 9600 bits of information.In order to correct for rotational inconsistencies 8 shirts left and right were required for each comparison of templates from CASIA-a data set.Now that optimum parameters have been determined,the FAR and FRR are 0.005%and 0.236%respectively,enough to applied to practical use;but the traditional Hamming distance 0.013%and 0.478%.