Scribd
Upload
26 views10 pages

Iris Recognition Using Wavelet Transform and SVMBasedApproach

Iris recogniton

Uploaded by

sushil
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
26 views10 pages

Iris Recognition Using Wavelet Transform and SVMBasedApproach

Iris recogniton

Uploaded by

sushil
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 10
Asian Journal of Convergence in Technology ISSN NO: 2350-1146 LF-5.11 ‘Volume V Issue I Iris Recognition Using Wavelet Transform and SVM BasedApproach Sushitkumar S. Salve Electronics and Telecommmnication Department Sinhgad College of Engineering. Pune, India sushil.472¢ _Abstract—Inthis paper we proposed an improved novel approach 1 identity the person using iris recognition technique. This ‘approach is based on fuszy and support vector machine (SVM) as ‘an iris pattern classifier Prior to clssiter, region of interest Le. iris region is segmented using Canny edge detector and Hough transform. Provided that the effect of eyelid and eyelashes get reduced. Daugman’s rubber sheet model used to get normatized iris 10 improve computational efficiency and proper dimensionatity. Further, discriminating feature sequence is obtained by feature extraction from segmented tris image using 1D Log Gabor waveletEncoding és done using phase quantization 10 get feature vectors, These binary sequence feature vectors are used io train SPM and fuzzy system as irs pattern classifier. The experimental {esis are performed over statdard CASIAIFSV# database. Keywords—iris, segmentation, Gobor fters, SVM,FCM Z IstRopuction Due to higher need of security in public as well as in personal information aspects, vigorousand reliable automatic recognition of individuals has been an attractive goal [1], [2} [5]{7].8]. In public security and information security domain biomettics technology plays an important role, Biometrics is standard of comparison based on either physical or behavioral traits, Fingerprint, iris pattem are physical traits while voice, signanure includes in behavioral traits. However, iris recognition is proven to be more beneficial as this accurately identifies each individual and also distinguishes one from another. The human iris is thin and omular region in eye ocated around pupil and covered by comea, as shown in figure 1, which able to provide unique and independent detail ‘information of individual [3]. Random texture of protected iris is mostly stable throughout the life. Iis patterns are too unique ‘even not only differ in between identical twins but also differ in between the right and left eye. Iris recognition requires ‘comparing a register or enrolled iris template against a newly amiail.com captured iris from eye of individual [9], [16]. The itis recognition is applicable in identification as well as in verification mode. In identification mode recognition system identifies a person from entire register template by searching a database for matching. Tdentification is “one to many’ matching style (18), [24], [29], [38]. In verification mode recognition system authenticate a petson's claimed identity from their's, previously enrolled patterns, ‘Verification is ‘one to one’ matching style igure 1 Human Eye Here, our objective is to overcome the challenge like high recognition rate of recognition system. IL ReLetep Work ‘Many researcher worked in the field of iris recognition, Daugman suggest first suecessfil| working iris recognition system [1]. Integro-differential operator uses first order detivatives of image intensity to locate circular edges of iris and upper and lower eyelids. By varying the radius r and center (.y) position of cireular contour the operator searches for circular path for maximum change in pixel values, To achieve precise localization, the operator is applied iteratively ‘with progressively reduced smoothing. In Daugman’s system, by quantizing local phase angle of imaginary and real patts of 1 Asian Journal of Convergence in Technology ISSN NO: 2350-1146 LF-5.11 filtered image, feature vector of 2048b were generated (1), [3] R. P. Wilde proposed a system which uses 4 level Laplacian to extract features of iis. After extracting features, normalized correlation and fisher classifier is used to clasify the features [11] Tsai, Lin and Tao used fizzy gray curve tracing method for isolate iris region. To extract local feature points and 0 evaluate dominant orientation of all detested feature points, multiresolution Gabor filters are used. Fuzzy matching algorithm is also introduce to calculate matching score [3 Wilde system compares feature point by using the information comprising the local feature and the position of each point [3 K, Seetharaman and R. Raghupathy adopted Canny edge detection method and circular Hough transfoma to seamen the iris [4]. Dangman’s rubber sheet model used to coavert polar coordinate to rectangular coordinate for fixed dimension ini [1], [4]. After decomposing 2D normalized iris to 1D signal feature will be extracted by convolving that ID signal with ID Log-Gabor filter using FFT and IFFT. These feature given to phase quantization process as input, to generate binary template ie. iris code, Hamming distance is used for bit-wise comparison for matching [1], [4]. Boles and Boashah ereaualyzed iris images by ze10 crossing of ID wavelet transform, Zero crossing representation is used to extract features. Different dissimilarity functionsare used to compare feanure vectors [10], Ma, Tan extract local details of iris pattern were capture by bank of spatial filter which are kemel based on Gaussian—Hennite moment and key local variation method using 1D wavelet transform [28]Sun adopted Gaussian filter to caleulate direction of itis image and then angle scale quantized into six discrete value. Yu uses ‘muitichanel Gabor filters to fetch key points of iris image. Related distances between these key points are used to represent fearures [3]. Hyun propased score level based iris recognition method using SVM (Support Vector Machine) and two Gabor wavelet filters (25) The remainder of this paper is organized as follows, Section IL describes the proposed approach for itis seamentation. An approach for feature extraction andan encodina is deseribed in Section IV. The proposed novel matching schemes by SVM and fuzzy system are presented in Section V and VI Sections respectively. The experimental results are shown in Section ‘VIL At last, Section VIII concludes the research work UL, Is Seovewzanos An eye image consist of iris texture and some irrelevant part as wel [13], [15], [12} [17] [23]. So, to isolate region of interest Le, annular iris region, the pupillary and limbic boundaries should be detected. inthis section, we describe, proposed edge detection and inner and outer circle detection using Canny edge detector and Hough transform [26] respectively. A Edge detection Every eye image is consist of different pixel values. Each, pixel is associated with its illumination. So, the abrupt changes wwwasianssr.org ‘Volume V Issue I in illumination are used by edge detector to get edge map of image. Here we proposed multi-step edge detection method uusingCanny edge detector. To apply Cammy edge detection, basic image processing aperation like smoothing, thresholding are used, The whole procedure is given below to apply Canny edge detection. 1, Tozeduce unwanted noise and texture Gaussian filter smooths the image with mathematical equation, 2(m,n)=G(m.n) * flm.n) Where, . oe 20° By using gradient operation, compute gradient of smnoothed image g(m.n). Mathematically, gradient of (un) computed as. M(\m,ni=\(gq:(m,n)+9,-(m,n)) 3. Thesholding [M\m,njif M(m.n'>T My\m,n\=f |x | 60. otherwise 4. Those nonzero Mr" are greater than their two adjacent along direction of gradient keep them unchanged, otherwise seit to 0. 5. Now set two threshold to get two binary images Ty. and TD 6. Then by linking edges in 7: and final edge map. ‘The edge map of input eye image using canny edge detector is shown in figure 2(b). B. Iris and pupil boundary detection ‘Now, the important task to find exact location of itis, itis also called iris localization. But it demands the radius and center co-ordinates of iris and pupil region. The standard computer vision algorithm, Hough transform plays an important role to determine parameter of pupil and iris. As iis and pupil are annular region, so we used circular Hough ‘wansform to get center co-ordinate and radius of iis and pupil so that, we get boundaries of respective regions. By taking gradient of intensity of eye image and then tuesholding the resultant image, we get an edge map of eye image. From an edge map. for the parameter of circle through Xe and we get an each edge point votes are cast in Hough space. Asian Journal of Convergence in Technology ISSN NO: 2350-1146 LF-5.11 Ye are the center coordinates and radius rare the parameter, by which we can define a circle according to equation, Ret Yer Using linear Hough transfomm eyelid can be isolate by fitting line to upper and lower eyelids. The resultant boundary detection and suppression [14] of noisy eyelid, eyelashes ar shown in figure 2(c) and figure 2(d) respectively. © Normatization ‘As we successfilly segment the region of interest, ie isolated ini region. Now we need to have fixed dimension its region 0 that we allow them for comparison. As illumination level gets vary the pupil dilation occurs. This causes stretching of inis and this is the reason for dimcusionally inappropriate iis ‘The sources like camera rotation, imaging distance may cause the inconsistency in iris images. So to produce the iris region with fixed dimensions, we need normalization due to which ive get two suaps of same irs in varying condition have the same important fearare at same spatial location. Daugman suggested a rubber sheet model which re-assign each point of (r.6) itis to polar coordinate where r belong to interval [o.1] and angle 9 ranges in [72 ]. The remapping is mathematically modeled as, /[*"8),y/"8) 1 1-r\x,/6/+rx,(8) yir,@=1-r|y,l6l+r y,(8) 1gxy) is iris image, (x.y) are original rectangular coordinates, (7-6) are corresponding polar coordinates. The *e x, and ¥p are coordinates of pupil, “i and Yi are coordinate of iris along angular directions ® . This model compensate the inconsistencies due to imaging distance and pupil dilation and produces constant dimension normalized iris patterns. The nonnalized output image is as shown in figure (0), wwwasianssrorg ‘Volume V Issue I Figure 2 Segmentation Reais (@ Original eye image (b) Edge map by cary edge detector (0) Afr applying Hough transform, (8) Eyelid supresion (c) Nermalizes ns region IV. FEATURE EXTRACTION AND ENCODING ‘To have accurate and precise recognition of person, the most useful and unique information about iris pattem must be extracted. For comparison purpose only significant information get encoded. Band pass decomposition (6), [17]. [23], [19], [20], [22] of ins image is famous in irs recognition technique to create biometric template. In this paper, nommalized iris pattem is convolved with ID Log-Gabor wavelet for actual implementation of feature extraction, Number of ID signal are recovered by breaking 2D nomnalized iris pattems. Then using FFT and IFFFT, convolution is done in between 1D signal and 1D Log-Gabor wavelet, Gabor wavelets are able 0 provide the best representation of signal in both space and spatial fequency. Sine/cosine wave get modulated with Gaussian to construct Gabor filter. Generally. a sine wave is much perfectly localizes in frequency than space. After sine modulated with Gaussian provide space localization but at same time it loses localization in frequency. The signal decomposition carried by Gabor filters. The drawback of Gabor filter is that, whenever the bandwidth exceeds one octave even symmetrical filter will hhave DC component. We can get the zeto DC component using Gabor filter for any bandwidth which is nothing but Gaussian over logarithmic scale, which is known as Log— 1 Asian Journal of Convergence in Technology ISSN NO: 2350-1146 LF-5.11 Gabor wavelet [11], (23], [24], [25], [26], [27], [28] filter. ts frequency respouse is given as, where, fo iscenterfiequency, % is filter bandwidth. ‘The feature extraction is cartied out by convolving 2D nomialized iris with ID Log Gabor wavelet. The 2D nonnalized iris is decomposed to 1D signal.1D signal is nothing but the row of 2D normalized ins pattern. Each row is corresponds fo circular ring on iris region, Let us assume a 4 * 4 matrix fiom normalized itis patter, 120125 150160 120 123156 166 125125155165 124 124154164] After convolution with 1D Log Gabor wavelet filter, the following complex number matrix is produced as feature —3.14-10,23— 1.44 12.82 1.75-i2.693.23+i0.17 —3.68 -10.27 —1.67 -i3.31 2.06-13.143.76+i0.21 —3.33 10.23 -1.58—-i3.00 1.861 2.923.48+i0.17 ~3.32 -10.18 ~ 1.611 2.971.82~i2.923.46+i0.14 ‘Volume V Issue I Figure 3 Phase quantization ‘The feature extracted using ID Log Gabor filter need to be encoded so that binary iris code will generate anc it helps in classification in matching scheme. Phase quantization is used here to generate iris code. As represents in figure 3, phase ‘quantization assigns 11 if both real and imaginary parts are positive, assigns 0,1 if real and imaginary parts ate negative andpositive respectively, assigns 1,0 if real and imaginary pparts are positive and negative respectively. and assigns 00 if ‘oth real and imaginary parts are negative. On the basis of this, logic binary stream get generated as shown in matrix form, ‘00001011 00001011 00001011 00001011, ‘This binary stream looks like as shown in figure 4, TP STAT Figure 4B steam code V. Mutictass SVMAS Ints PATTERN CLASSIFIERS ‘The code extracted from input iris image need to compare with previously enrolled iris code. This comparison is done based ‘on Support vector machine (SVM) classifier (21], [31]. [33]. [34], [36], [39]. By finding decision surface between two classes of input vectors from training set. As shown in figure 5, SVM tries to maximize the distance of separating boundary of two classes by maximizing distance of separating plane [32] of each of feature vector. Figue 5: SVM cassie fx is input vector then a(x) is linear fanction given as, gx =w.xeb Asian Journal of Convergence in Technology ISSN NO: 2350-1146 LF-5.11 ‘2(8) represents straight line if x is two dimensional feature vector, Ifit is 3 dimensional then a(x) represents plane, Now, x is greater than 3 dimension then a(x) represents hyperplane as shown in figure 5. Here, w is weight term which is perpendicular to hyper-plame, Orientation of hyperplane is given by w. Also, b is bias term which indicates position of hyper-plane. Hyper-plane divides D dimensional space ito two half spaces. In two class problem classification rule if Gixl=w.x+b70 then, x belong to class ci. IF 9.x)/=w.X+D<0 then, x belongs to class cs. While classifying any input vector we assign class belongingness y= flap NEC: then Yi = #1 andi¢ XC |w.x,+b)>0 then Ji Hence Yi! is always positive whether *! belongs to ©: or ©2 . Now for stability of each feature vector we considera margin |Y! Therefore, wextb>Y ris marin is nothing but a measur of x, distance of *i from hyper-plane. So, distance of hyper- wath plane fiom that point *' is given by “Th 2Y By proper scaling we can set YollWI= therefore ‘wx+b=1 Now classification mile becomes. W.xtb>1 > x, €c,Aw.x#bS1 > x,€c, até is support vector then Ys ¢w. 1 +b-1)-To design SVM we uced margin as large as possible. So, margin w.xeh will be TIw]] -Y To get largest possible margin two wll @ point take into consideration. (1) Minimize maximize b. This problem can be convert to un-constraint ‘optimization problem by using Lagrangian multiplier L\w,b,@ $l PY al yxy Where, is Lagrangian multiplier. To optimize that take partial derivatives of above Lagrangian with respect to wand D. So we get aL. oL BLS yt aaa ob ‘Volume V Issue I Finally, we get 1X a-F LD @.a,y.v)(%,) a a, @, To, become SV must be very large, It means. shows the effect of feature vector on decision boundary. In order to obtain a non-linear decision boundary, we replace the ier product [%*%j) with a nonlinear kere! K{[x,.x,Aget fix=Doy,a,K|x.x,)+b ‘The simple idea behind the nonlinear SVM [5] is to use a Kix.x) ‘kernel function to map the data x from the input space to the new higher dimensional feature space on which ‘the mapped data points become linearly separable. The hidden ‘units on the hidden layer are characterized by the kemel function adopted by SVM. The two basie kemels used in this work are, Radial Basis Function (RBE) kernel: 2 ‘VI. Fozzy pase Ikis PATTERN CLASSIFICATION 2X, 0 Polynomial kemel: K(x.x,} [16 Fuzzy classification is process of grouping an individual having same characteristics into fuzzy set. We use Fuzzy C- ‘Means (FCM) algorithm [3] for fuzzy classification of iis pattemsFuzzy C-means is a method of clustering which allows one data point to belong to one or more clusters. Fuzzy patttioning is carried out through an iterative optimization of the objective function shown below, with the update of membership “v and the cluster centers ©) by, Asian Journal of Convergence in Technology ISSN NO: 2350-1146 LF-* where, [Cf] is distance fom point 1 to cument ctuste’s center, [MTG] is tie stance fom i point 1 to other cluster centers ‘The iteration will stop when, max; |u*!—u"|

You might also like