Research Of Iris Recognition Algorithm Based On Fractal Theory

Information age brings convenience to people, but at the same time brings new challenges to personal identification. Traditional technologies based on smart cards or passwords have been out of date. Biometrics provides a reliable solution for personal identification and is becoming more and more important. Biometrics deals with identification of individuals based on their biological or behavioral characteristics. Biological characteristics are innate, such as fingerprint, face, iris, etc. while behavioral characteristics are acquired, such as handwriting, gait, etc. Iris Recognition has great potential because it's highly accurate, noninvasive and highly secure. An Iris Recognition System consists of four modules which are image acquisition, preprocessing, feature extraction and matching and has two operating modes. In identification mode, matching is one to many comparison while in verification mode, matching is one to one comparison. Based on recent advancement in iris recognition and standard techniques from image processing and pattern recognition, this paper discusses issues about preprocessing, feature extraction and matching. Some methods are proposed and experiments are also described. Specially designed hardware required in image acquisition is not studied in this paper. Instead, CASIA iris image database 1.0, provided by Lab of Pattern Recognition, Institute of Automation, Chinese Academy of Science, is used. It includes 108 eyes and each eye has 7 images. All experiments are based on the database in this paper. Preprocessing is composed of three steps which are iris location, normalization and image enhancement. Location time usually account for about half of the entire iris recognition process. So to be real-time, how to locate iris rapidly and accurately is a key step in preprocessing. In this paper the method used in coarse location of pupil is improved. First, binary the iris image. Then, estimate the center and radius of the pupil roughly. This method raises the performance of coarse location and decrease the searching range of fine location. The inner boundary of the iris is located by using the circle edge detection algorithm in a binary image, which avoids the problem caused by bright spots with the pupil. The outer boundary of the iris is located by using Canny edge detection and Hough transform algorithms, which can get satisfactory result even if the iris texture is very clear. The Experiment shows, the iris location algorithm proposed in this paper, reduces location time greatly on the premise of high accuracy. To make iris recognition shift, scale and rotation invariant, the iris image is normalized into a rectangle whose size is 64×512 in polar coordinates. In order to reduce the impact of uneven illumination, local histogram equalization is done in the normalized iris image.There is a natural relation between fractal and images. Fractals generated by a simple iteration can be used to simulate natural scenes and the complexity of natural scenes can be measured by fractal dimension. This relation is why fractal theory can be applied to image processing. Fractal dimension is also called Hausdorff dimension and is an important basis of fractal theory. It popularized dimension from integer to fraction and is fit for any set. It is very difficult and even impossible to compute fractal dimension of images according to the definition. There are many approximate methods and three classical methods which are called spectrum dimension, blanket dimension and DBC (Differential Box-Counting) dimension, are implemented in this paper.Fractal dimension and the perception of human vision with coarseness of images is consistent and because taking into account different scales it's also anti-noise, so fractal dimension is a very good representation of texture feature of images. Any method of computing fractal dimension can be used to extract texture feature of the iris, but different methods have different scopes of application. Also different textures may have the same fractal dimension and this may be due to combined difference in coarseness and directionality (dominant orientation and degree of anisotropy). To solve the above problems, a new method of feature extraction is proposed in this paper. First, move a window in the iris image by some step length and get 32 iris subimages. Then, four additional images which are the high gray-valued image, the low gray-valued image, the horizontally smoothed image and the vertically smoothed image, are computed from each iris subimage. Last, compute spectrum dimension, blanket dimension and DBC dimension of the origin image and four additional images as features. According to distribution characteristics of iris texture, the classifier based on Manhattan distance is improved. Give different weights to features from different iris subimages to reflect the confidence level. The Experiment shows, the method proposed in this paper brings down EER to only 1.24%.Based on preprocessing, feature extraction and matching Algorithms implemented above, an iris recognition system is developed in this paper. The system has five modules which are login, register, search, access and manage. It's only a prototype system and simulates the basic functions of commercial iris recognition systems.The algorithms proposed in this paper have got a good result, but to be commercial, still need further improvement. In the future, evaluating the quality of iris images and using a more effective classifier based on machine learning algorithms should be considered.