Research On Iris Feature Extraction

Iris identification technology, as one of the biometric features recognition, its drawing high level of attention has run over the years, the reason is that its object recognition, iris, the unique attributes in constitute of human organs, make it become the most popular one of the modern security identification technology. Extracting feature in the whole process of the iris identification is a especially significant part, as a result, this article will spread the discuss on the basis of it. This article selects the classic Gabor filter group, cooperating the principal composition analysis method to Extract characteristics and make classification of the iris.Firstly, this article introduces a series of steps as well as preprocessing algorithm of the iris image, acquiring the normalized iris images by quality assessment, hough algorithm locating, polar coordinates conversion, the ambient light supplement. This article studied the 2D-Gabor filter group, discussed the meaning of each parameter according to the Forming principle and formula of the group, and conducted filtering experiments in allusion to different values of the scale, orientation, frequency of the group, observing the filtering response.Then the paper analyzed the effect of different parameter selection on the filtering results, thus divided value of these parameters on the rationality, finally acquired the most efficient information after filtering.Secondly, this article studied the Principal Component Analysis deeply, its idea is to explore how a small number of several main composition characterization represent the internal distribution of the large amounts of data, through a certain method to extract the small amounts of some main parts from the initial variable, make these parts retained the maximization of the original variables’ characters, ensuring they are independent of each other as well. The realization was by the principle of linear algebra in mathematics, turning the original variable to the linear combination,mapping to another space as the new indicator. This article merged the data after filtering through the Gabor filter group, and formed high dimensional information matrix, then executed according to the dimension reduction processing steps, andobtained the low dimensional characteristic matrix from samples of the same kind as last iris feature for comparison.The article also introduced the distance, a method to describe similarity, and illustrated several kinds of common distance as the dependence of comparative test.At last, the experiment used the data of 345 samples of 39 classes from the CASIA_V1、the CASIA_V3 and the JLU_V2 iris databases, choosing different kind of distance, and conducted tests according to the above process. Analyzed the distribution of Within-Class and Between-Class, the discrimination, and the correct recognition rate between the training set and testing set. As the experimental results, the recognition system has a good effect, proving the the feasibility of the method, but it still needs further study to improve in some aspects.