Iris Recognition Based On Log Gabor Transforming And DLPP

To meet the increasing security requirement of the society, people pay more and moreattention to personal identification. Traditional methods for personal identification, such asID cards, keys or passwords, however, are usually not reliable. Therefore, a new method forpersonal identification has been attracting more and more attention. The human iris is theannular part between pupil and sclera with stable and distinct characteristic. It is the iris thathas brought a revolution in personal identification.Currently, the iris has achieved the highest recognition rate in all single featurebiometric systems. What’s more, the iris recognition systems have the advantage ofnon-contact image acquisition, rapid recognition, that’s enough to make an importantresearch direction. In general, the process of iris recognition including five steps, imageacquisition, iris localization, feature extraction, feature matching and decision. This paperhas made depth research in iris recognition, and proposed an improved algorithm based onthis process.First, we proposed an improved iris localization algorithm. Iris localization is a keystep in iris recognition. As literature review’s algorithms are relatively time-consuming andnot enough robustness, this paper proposed the iris localization algorithm based on waveletdenoising and Hough transform. Iris have the occlusion of the upper and lower eyelids andeyelashes, pupil surrounded by reflected light spot, all of these noises make iris localizationmore difficult. After wavelet tranforming of iris by using soft threshold filter，we can get ridof the high frequency components and retain the low frequency components. So the mainpart of the iris is reserved and the noise part is discarded. Then, we use the Houghtransforming method to locate the inner circle within the binarization iris image.The invalidregion is removed by the use of located inner circle which is helpful for extracting the outercircle. Experiments on the CASIA iris databases show that this method not only improvethe localization accuracy, but also the localization speed.Then, we proposed an improved iris normalization algorithm based on the rubber sheetmodel proposed by Daugman. We sampling pixels along the radial direction of iris all ofwhich are then converted from polar coordinates to Cartesian coordinate. After that,different annular irises have the same rectangular shape, the normalized iris bars are finallygot which are important for further processing.At the recognition step, we proposed a combination of frequency domaintransformation and spatial characteristic extracting algorithm. First, we make a comparisonbetween Gabor Filter and Log Gabor Filter. Then, we use the4scales and8directions LogGabor filter to transform the normalized iris and get32frequency-domain iris images. We combine these32images into a big Log Gabor iris to take place of the original one.Obviously, the Log Gabor iris is too big to compute, as is32times bigger than the originaliris. So dimension reduction must be operated. We use3times of wavelet decomposition onthe Log Gabor iris and extract the LL3level sub-image which is1/64times of the LogGabor iris. After wavelet decomposition, the dimension of the iris is suitable for featureextracting. We have derived the LPP and DLPP algorithms in detail and made a comparisonof them. The DLPP algorithm is the improvement of the LPP algorithm which considerboth the local structure information and the global discriminate information that is benefitto recognition. We use these two algorithms to extract the feature of iris, feature matchingand decision are made following.Finally, lots of comparative experiments are done on the CASIA Iris database.Experiments results show that by the use of Log Gabor transforming, the recognition ratehas a significant improvement, nearly about10%. At the same time, we have compare thePCA, LDA, LPP, DLPP algorithms. Experiments results show that DLPP has the idearecognition rate, as it can reach98%when the training sample is relatively large.In summary, this paper has made a detail research on iris recognition. Proposed animproved iris localization method which has good performance in both localizationaccuracy and speed. At the feature extraction stage, proposed an iris recognition algorithmbased on the Log Gabor transform and the DLPP. Lots of experiments have done on theCAISA iris database which verified the feasibility and effectiveness of the proposedalgorithm.