| A fire-new demand of social security has been put forward because of thedevelopment of society. In this condition, traditional security technology presents a bigdisadvantage. A convenience, valid, secure personal identification technology is neededurgently. Biometric technology is an important method to solve the problem of socialsecurity.Biometric technology of human being can realize personal identificationautomatically by use of image processing and pattern recognition based on the inherentphysical or behavioral characteristics of human being. The biological feature of humanbeing includes voice, fingerprint, lines of palm, iris, eyeground, facial features, DNA,signature, handwriting, etc. The kernel technology of biometric lies in these aspects:how to obtain these biological features, translate them into digital information, storethem in the computer, realize personal identification use credible matching-algorithm.Among the different biological features, iris is one of the most strong andpermanent feature. As a feature of identification, iris has more excellent characterscompared with other features, such as exclusivity, stability, collection, non-invasive. Inthe recognition of biological feature, non-invasive is a necessary development directionof the search and application of personal identification. Compared with thosenon-touched recognition methods, such as facial recognition and voice recognition, irisrecognition has higher accuracy. According to statistical data, the inaccuracy of irisrecognition is lowest among all kinds of biological feature recognition methods.Personal identification technology based on iris recognition gets more and moreattention of academe and enterprise. The system based on iris recognition will has moreaccuracy. Because of the influence of collecting equipment, illumination, translation,rotation, scale variant in the process of iris imaging, iris recognition is full of challenge.In this paper, the important function of biometric technology, the capacity, utilityand development of iris recognition system, and some typical algorithm are discussed indetail. The emphases and difficulty in iris recognition are also pointed.Iris recognition is a kind of pattern recognition. In this paper, some new recognitionmethods are generalized comprehensively, the implement process of iris recognitionsystem is discussed in detail. The full system is composed of iris image obtaining, irisimage preprocessing (including iris location, normalization, image emphasizing), irisfeatures extraction and encoding, iris registering, feature database building, patternmatching, classifier designing, and so on.Other researcher's works are analyzed in this paper. Combined with author's ownresearch, iris recognition technology is studied profoundly from three thesis includingiris image preprocessing, image features extraction and encoding, pattern matching andclassifier designing. According to the actual need of iris recognition technology, someimproved algorithms are given focused on the emphasis and difficulty in this field, someavailable image processing methods are attempted and result is acceptable. These will bediscussed as follows:1. Some processing skills and theories on Daugman iris location method areanalyzed. The iris location algorithm is improved in order to improve the speed andaccuracy. The improved algorithm can locate the center and radius of the pupilaccording to the gray scale of sclera, iris and pupil, then for the boundary of sclera andiris disturbed by complicated noises, an Edge Detection Threshold Analysis Method(EDTAM) based on statistical principle is given to distill the external boundary pixels ofiris, minimum ercheng is used to get the center and radius of the iris.After distill a big grads region got by statistic function, EDTAM extracts theexternal boundary pixels by non-maxima suppression algorithm. In this paper, before thethreshold is selected, each pixel grads is local standardized, then fuzzy part and ill-suitedthreshold are deleted. The statistical character makes the input parameters haverobustness to the disturbed image. The result shows that the algorithm has better stabilityand robustness compared with typical boundary detecting algorithms.2. Conventional iris feature extracting and encoding methods are discussed andanalyzed in detail. Three improved algorithms are given as follows:(1) 2D Wavelet Transform combined with Integral Image;The normalize iris image is demodulated by using method of 2D wavelet transformcombined with integral image, then get the features vectors for the purpose of iris texturefeatures extracting.(2) Zero Spectral Moment Filters (ZSMF);ZSMF is a new kind of filter, is has good performance and it is better than FIR(Finite Impulse Response) which is used in conventional signal processing. In this paper,a new iris texture feature extracting algorithm is given based on research on ZSMF. Thisalgorithm demodulates preprocessed iris image by using balanced ZSMF. It can extractmore accurate iris texture features combined with integral image.(3) 1D Features Extracting;The algorithm is very simple. It locally proportions iris textures by using LTI, thenget the iris textures features whose gray scale invariant. It can avoid the effect ofillumination variations.3. Method of pattern matching and classifier designing are studied and a newclassifier is given. That is Spectral Information Divergence and Spectral AngleMapper(SIDASAM) classifier.SIDASAM classifier takes advantage of SID and SAM and it has good classifiedeffect. Under one-to-many condition, iris images from CASIA are chosen for test. Dataof the first five images which have the lowest SAS value is given in a table.At the final of the paper, the implement of algorithms mentioned above is describedin which data come from CASIA iris image database. The result is also given.It is shown from theory and practical test that the improved algorithms given in thispaper are better than traditional iris recognition methods. At the same time, the methodscan avoid the effect of illumination, translation, rotation, scale variant, and have betterstability and robustness.Iris image processing technology has rich content and applied foreground. Woks inthis paper are only a little in this field. There are many works to be done. The author willmake great efforts to improve the algorithms or give new algorithms and test them withlarge amount samples to validate the haleness of them.Although iris recognition technology is presented from 1900's, people didn't godeep into researches on the algorithms because of hardware condition. So it is notapplied in a large scale. But today, hardware has developed rapidly. The capacity of theproduct is higher and higher while the price is lower and lower. And mathematics andcomputer science are developed. All of these sustain the development and application ofbiological feature recognition technology. I believe it will be faster and faster totransform the theory of biological recognition technology into production. Theproduction will have more capacity and lower cost. Among these, iris recognitiontechnology will be a superexcellent one due to its development and mature and theadvantages of iris.
|
PDF Full Text Request