Research On Iris Recognition Technology Based Multiwavelet Transform

Iris biometrics characteristic has some good advantages, such as stability,unique, and it can't copy and it can't infringe, then the biometric features can be usedto identify the person in high-safety area. With the safety consciousness boosting up,especially from "911", iris recognition becomes one of topic research filed in biometricrecognition. Multiresolution analysis provides powerful facility for signals analysis andit can get signals optimization time-frequency characteristic. Meanwhile, as a newmethod, multiwavelet technology has some better merits than wavelet and waveletpackets, and it can apply in iris recognition process to achieve valid iris features'multiscale representation. In this paper, it introduced iris serial image qualityassessment, iris fast localization, multiwavelet feature extraction and iris featurematching methods, and the aim is to obtain the representation for iris multiscalefeature and the method of iris feature matching and improve iris recognition rate viamultiscale analysis and iris geometry topology analysis.As usual, iris image quality affects the results of iris recognition directly, and thehigher quality iris image should be selected during iris automatic sampling.Accordingly, the iris image quality assessment is the base of iris automatismautomatic collection. In generally, human visual characteristic has some traits, suchas feature selectivity, interesting region selectivity etc., and the wavelet transform hasa better time-frequency property. So the human eyes property is modeled by wavelettransform in the paper to assess the image quality. At first, the interesting region islocated by iris localization, and then the region is decomposed by wavelet transformto obtain the special sub-bank signals. As usual, zero-crossing points reflect signalfeatures, and wavelet zero-crossing points are extracted to get interesting region'scharacteristic points and to implement iris image quality assessment. The methodimplemented effectually iris serial images quality assessment and reached to theneed of practical application.Iris localization is an importance process in iris recognition, and it influencesdirectly the performance of iris recognition. Currently, the mainly iris localizationalgorithm includes integral-different operator and Hough transform, but the irislocalization algorithm has some drawbacks, such as the influence of eyelid andeyelashes and localization time is long, and these disadvantages influence real-timeperformance in iris recognition. As usual, the voting method is important to detect themodel accurately, and then the fast iris localization algorithm is proposed by theprecondition voting method. In order to reduce the redundant voting and thecomputation complexity, the edge points is sampled to complete the sample set.Three points are selected in the sample set to confirm the candidate circle, and thecandidate circle is voting to find the optimization candidate circle as the detectingcircle. In order to reduce the effect of eyelids, eyelashes, the coarse-to-fine strategy isused to localize the iris fast, namely, the coarse pupil center is estimated firstly, thenthe fine circle edge is localized in region of interesting. In experiment, the proposedmethod is compared with the integral-different operator and Hough transform. Fromthe results, it showed the proposed method has better performance than others, andthe method is fast and robust.The multiwavelet has the superiority than the wavelet and wavelet packets, and ithas the better local property and spatio-temporal orientation selectivity, this iseffective to analysis mutation and oddity signal. As a same time, the base function ofmultiwavelet has the different selectivity, and is easily to obtain the tight support ofsignal. Generally speaking, the higher frequency component is easily to interrupt bynoise, and the detail feature is gathered in low frequency components. To GHMmulti-wavelet, the energy at most is gathered in low frequency components. The irisimage is analyzed after the multiwavelet transform to get the special subbanks in thelow frequency components and then the feature is represented in the specialsubbanks to obtain the robust features. The zero-crossing points is reflect themeaning feature in signal, and it is not easily to effect by noise. Then the iris featuresare represented by zero-crossing point in the special subbanks, and therepresentation is robust and fast.Iris feature matching is an important step in iris recognition, and the aim is to findthe optimization covering in the feature space to classify the iris, namely, it searchesthe least distinguishing distance in the class and the most distinguishing distanceamong classes. Since the feature space is vector space by the multiwaveletdecomposed, the vector distance is to be used to classify the iris. In the paper, theangle of vectors is used to judge the difference between the vectors, that is to say, thebigger of the vector angle, the bigger difference of the vector. In order to normalize thedistance of the judgement, the cosine distance is used to classify the iris in thehigh-dimension vector space. In experiment, in order to select the special sub-bankseffectively as feature space, the different sub-banks are test under differentillumination, noise circumstance, and the two special sub-banks in low frequencycomponents, which are not effected easily by vary of noise and illumination, areselected as feature space. At the same time, the optimization judgement threshold isconfirmed by the statistic analysis between the same class and different classsamples. In additional, the hamming distance is compared with the cosine distance inthe experiment. From the results, it showed the method is effective and robust.