The Research Of Iris Recognition Algorithm

Following the Internet and information technology development,traditional identity recognition way based on password have more and moredifficulty to satisfy those high security required occupation. Biometricrecognition technique become one of the hottest research fields, which makesuseoftheinherent physiological orbehavioral characteristics ofhuman beingsto identify individuals. Due to the unique, stable and live physiologicalproperties of the iris and non-invasive to users, high reliability in practicalapplications of iris-based system, Iris recognition is considered as the mostpotentialtechnologyinbiometrics.Iris recognition technology integrates mathematic, computer science,optics, electronics and physiology etc. As an application-oriented researchproject, iris recognition has achieved great deal of progress in the past decade,but manyefforts remain to be taken to further improve the performance of irisrecognition systems. Aiming to develop an iris recognition system andevaluateour proposed algorithms,weinvestigate someofthekeyissuerelatedto iris recognition in this thesis, including iris image pre-processing, irisfeatureextractionandmatchingalgorithms.In general, iris image pre-processing consists of iris location, eyeliddetection, normalization and enhancement. Location time usually account forabout half of the entire iris recognition process, and become the main factoraffectingthereal-timeofthesystem.Sohowtoquicklyandaccuratelyfindtheirisareaisacrucialstepinirispre-processing.Wefirstintroducetwoclassicallocation algorithms, and then we propose our improved algorithm based onthem.Taking into consideration the fact that interior to pupil, there would havesome lighter spots because of reflection. This paper improves the commonlyused coarse location method. It utilizing the gray scale histogram of irisgraphics, first compute the binary threshold, averaging the center of chords tocoarsely estimate the center and radius of the pupil. After that, the innerboundary is fine located using the algorithm of circle detection in the binarygraphic.Thismethodcouldreducetheerroroflocatingwithinthepupil.Then,this paper combines Canny edge detector and Hough voting mechanism tolocate the outer boundary. As we have known the parameters of inner circle,thespeedofHoughtransformationisincreased.Based on CASIA iris database, experiments among proposed algorithmand two classical experiments showed that it could greatly decrease thecomputationalcostwithoutreducingtheaccuracy.Inaddition,astatisticalmethodisexploitedtosearchandexcludeeyelashand eyelid areas. Experiments have shown the applicability and effectivity ofthisalgorithm.Gabor filters are band pass filters which are used in image processing forfeature extraction and texture analysis. It has obviously optional character inorientation and frequency and can acquire optimal co-resolution in bothSpatial and frequency field. It was shown by several researchers that theprofile of simple-cell receptive fields in the mammalian cortex can bedescribed by oriented two-dimensional Gabor functions. So 2-D Gaborwavelet filter can get a nice effect in texture recognition and segmentation. Ithasbecomeapowerfultoolinpatternrecognitionfield.For iris feature extraction, we encode iris feature based on Gabortransform. First of all, we bring forward a new method to divide iris textureinto sub-block: By our experiments in CASIA data base, two importantfindings can be summarized. In rectangular iris images, the most top 75% ofthe region contains most separable information; upper and lower eyelid blockdifferent size of the iris. The results of non-iris region detection can beemployed to divide the usable iris region. Through the calculation of therollingangle,theerrorintherecognitionprocess causedbyeye'sturningcouldbeeliminated.When encoding, besides the traditional way by Daugman, the paperintroduce two methods based on Gabor filter: even part of Gabor filter anddiscrete Gabor filter.To theformer, we use16 filters withdifferent parameterson each usable iris sub-block. And then, the absolute deviation of everysub-image as the iris feature for iris code. To the latter, we employ phaseinformationofthereal andimagepart afterfilteringforencoding.The codeofsub-blockinthenon-irsareasetzero.In the last step, pattern match, we make use of Euclidean distance toclassifythecodebasedonevenpartofGaborfilter,andHammingdistancefordiscrete Gabor filter. In CASIA data base, experiments in authentication andrecognition pattern have been taken to evaluate our proposed method offeature extraction and matching. The ROC curve shows that the result of ouralgorithmhasbetterrecognitionability.Allinall,thispaperestablishedanintegratediris recognitionsystemwithCASIA as the image database, and made some useful exploration in the keyissues of Iris recognition such as feature extraction and image preprocessing,andmadesomeresearchachievements.