The Research Of Iris Recognition Based On The Biomimetic Pattern

Biometrics recognition technology has been in harmoney with our daily lives and play an increasingly important role with the development of one high degree of modern information society. One type of Biometrics recognition technology--- iris recognition technology ---with its own merits, for example, incomparable uniqueness, stability, reliability, security and so on decides its stature of identity authentication technology that can not be replaced , and iris recognition technology has been focus in the research region of biometric technology .With iris recognition technology gradually matures, Continuously improves performance and lowers prices, iris recognition will has broad application prospects and broad market potential in areas such as financial securities, information, defense, transportation, public security , justice and daily life.The first part of the paper introduced the development process of iris recognition and research status, and introduced the iris recognition system, summarized research achievement of the iris recognition. The paper pointed out that at present the classification methods of iris recognition mainly adopted traditional pattern recognition methods, on which basis pointed out the limitations of traditional pattern recognition that in the face of the first N+1 sample, it might identify incorrectly the first N+1 sample as N-type sample. This elicited the bionic pattern recognition theory, started from "recognize" the concepts, aiming at the best coverage of feature space distribution of a type of sample. It recognized the samples group by category, and could be able to better deal with the fist N+l sample problems.The focus of the study for biomimetic pattern recognition is the construction of the same type of sample object in the feature space covered by high-dimensional geometric. In practical applications, it should be targeted at the different needs using different methods. According to the distance mapping of the same iris samples concentrated distribution near the centroid of iris, and the distance mapping of the heterogeneous samples is farther than the same iris samples, we assume that the geometric region coveres the same iris samples is a closed hypersphere. In the guidance of biomimetic pattern recognition theory, the paper adopted K-Means clustering algorithm to obtain the centroid of the iris samples, according to the distance mapping distribution of vector to get coverage radius. By this means, the paper constructed the coverage geometry region of the same iris sample images in the feature space, and then recognized the iris samples through identifying whether the pending iris subband was in the corresponding coverage area.This paper does much research in the following areas :1) At the first stage of feature extraction, this experiment based wavelet decomposition transform of the normalized size of 512×64 iris images , selecting the decomposed low-frequency band LL₁, treatment for feature extraction, using subbandlocal point iris the direction angle to indicate the characteristics of the point of the iris, with the direction of x and y express the partial derivative at (x, y) of the direction angle. The experimental results shows that the method used in this paper can effectively extract iris feature,and get better effect of iris recognition.2) In this paper, 10 types of iris images are randomly selected from JLU-IRIS iris.6 categories of the iris are regarded as the first sample set and the remaining 4 categories as a second sample set. When identifying, 300.400 samples were selected as the training samples from each type of the first sample set. Under the selected parameters and with method mentioned in this article, constructed the geometric structure of space covered by the regional characteristics of each type iris. Experimental results shows that when the sample size is 300 ,the recognition rate is 95.17 percent, the error recognition rate is 1.11 percent, the error refusal rate is 3.72 percent; and the size is 400 the recognition rate is 96.50 percent, the error recognition rate is 1.041 percent, the error refusal rate is 2.458 percent; It can be drawn from the exeperiment that large sample volume is better to obtain the effective result of iris recognition, because with the increace of the number of training iris samples, the more related information is provided and the more detailed coverage of the region was formatted, can be widely used in practice.3) Under the above-mentioned experimental conditions, compared with the methods of the cosine distance classifier, SVM classifier traditional iris recognition with the same parameters, test iris recognition and compare the identification results. the identification rate of method mentioned in this paper is 96.50 percent. The cosine distance classifier recognition rate is 92.58 percent. SVM classifier traditional iris recognition rate is 97.5 percent. The experimental results show that the iris recognition method based on biomimetic pattern recognition used in this article can obtain a higher recognition rate than traditional pattern recognition methods Under the same conditions. Can be widely used in iris recognition4) At the basis of geometric coveraged region of feature space have been constructed in experiment 2 of this paper, identification test has done for 1,600( 4×400) times relying on the second training samples that are not trained. Experimental results shows that the right rejection rate is 98.56 percent. higher by the method used in this article to the iris sample not participated in the training. This method resolves iris recognition category N +1 problem well, and effectively ensures the security of iris recognition demand, effectively prevent the invasion of illegal authority staff, has a strong practical significance.Biomimetic Pattern Recognition is under the guidance of Similar samples of homology between the a priori probability of the continuity of the law, research on the distribution of sample points in high-dimensional space, achieve the coverage of complex geometry. How to construct the coverage of complex geometry high-dimensional space is urgent problem need to be solution in biomimetic model of iris recognition, the scientific researchers should focus on exploration, more important is that provides a new direction of development for the field of pattern recognition.Experiments are carried out only at large sample size with the Identification methods mentioned in this article, for small sample size experimental support is still a lack . Construction of the iris characteristics geometric cover is difficult under the condition of large sample size. So at a low sample size constructing covered region of characteristics of the iris is the focus of future study.Although the experiment made shows better identity effect, but because of time constraints and limited ability, a lot of work should be done for improvement and perfection, such as doing some improvements on selection of iris sub-band--selecting with wavelet decomposition; obtaining better results after filtering feature extraction that iis conducive to the subsequent identification; On this basis extracting characteristics formatting of different sub-band structure of the geometric coverage area, and integrating recognition results of each sub-band to make decision, to obtain better results.The application prospect of biomimetic pattern recognition is obvious to all, it has rationality and universality. With the gradually improved and maturity of biomimetic pattern recognition theory, the high-dimensional space geometry coverage theory will obtain more fruitful achievements in the field of information, science and other areas.In the future studies, we can integrate the advantages and disadvantages of traditional pattern recognition and the biomimetic pattern recognition, consider to combining them effectively, and perfect and develop pattern recognition theory in its guidance to advance the process of pattern recognition. This is also the powerful guarantee of the iris recognition be widely applied. It needs subsequent efforts to take iris recognition application to numerous households to serve the production and living of human better, strive early for achieving marketization of iris products.