Real-Time Iris Recognition System Using A Proposed Method

Size: px

Start display at page:

Download "Real-Time Iris Recognition System Using A Proposed Method"

Oliver Mosley
6 years ago
Views:

1 Real-Time Iris Recognition System Using A Proposed Method Eri Prasetyo Wibowo Gunadarma University Indonesia eri@staff.gunadarma.ac.id Wisnu Sukma maulana Gunadarma University Indonesia wisnu sm@student.gunadarma.ac.id Abstract Iris Recognition has been learnt by many researchers since it can be used to detect and recognize someone better in biometric system. Hence, many researcher have been trying to improve the algorithm for iris recognition self. But, the most problem happened in doing the research is to do the iris localization well. Besides that, the eyelids and eyelashes are also the another problems in iris recognition since they could cover the iris or eye, and it might be some noise affecting in iris image as well. In this paper, we introduce a proposed method and try to implement it into real-time system to solve the problems. Keywords-Iris; Recognition; Localization; Approach Algorithm; Real-Time System. I. INTRODUCTION Iris recognition has been used for many decades and shown to be a highly accurate method identifying people by using the unique patterns of the human iris among the other biometric system[9]. Human iris on the other hand as an internal organ of the eye and as well protected from the external environment. It is the annular part between pupil and sclera, and has distinct characteristics such as freckles, coronas, stripes, furrows, crypts, and so on. It is an inner organ visible outside; hence, iris image can be captured without physical touch. Each eye has its own iris pattern that is stable throughout ones life, yet it is a perfect biometric for an identification system with the ease of speed, reliability, and automation. The human eye structures is shown in figure I. FIG. 1. Human Eye Structures Iris recognition efficacy is rarely impeded by glasses or contact lenses. Iris technology has the smallest outlier (those who cannot use/enroll) group of all biometric technologies. The only biometric authentication technology designed for use in a one-to many search environment, a key advantage of iris recognition is its stability, or template longevity as, barring trauma, a single enrollment can last a lifetime. Effort to devise reliable mechanical means for biometric personal identification have a long and colorful history. However, Iris recognition for identifying person was originally proposed by Frank Burch, MD in 1936[2]. The basic idea of its is to use the iris code that resulted from iris algorithm. FIG. 2. Iris Recognition Flow Diagram As shown in figure 2, iris algorithm or iris recognition method commonly consists of three main modules as following: Image acquisition Image Acquisition is the first step of all process in iris recognition. The aim is to acquire the image of eye. This can be done by using any tools like CCD camera, Infrared camera, etc. However, the most important thing is to get the good image that can be process into iris recognition system well. Pre-Processing and Feature Extraction Some researchers argued that the iris algorithm is the pre-processing part. Since the pre-processing includes iris localization, feature extraction, encoding. Matching Matching refers to the final decision for recognition or identification person. This isn t only matching and comparing iris, but also getting the information from iris. In matching system, it can be tested by using the matching algorithm such as Hamming Distance, Euclidean, etc. II. RELATED WORK With the increasing interests in iris recognition, more researchers devote their attention into this field. Now, researchers have done research in iris recognition majority using the Daugman s and Wildes method. In 1993[4], Daugman built a method that used integral differential operator as circle search operator to search over the image domain for the maximum with respect to increasing

2 radius, of the normalized contour integral along a circular arc of radius and center coordinates. And then, it would make 256 byte length of iris code with using 2D Gabor Wavelet filter to find Hamming distance in matching the similarity from two iris code. But, it has some weaknesses. The main of the weakness is in the calculation of blurring factor that is too sensitive toward the light reflection and contrast. And, the calculation of the circle equation to get the iris area is dependent to the radius and middle point parameter. In 2007, Daugman made new method with active contour approach in segmenting the area of iris, and fourier method to process the image with any direction of eyes to the camera[5]. Otherwise, Wildes method converted the eye image into a binary edge-map via gradient-based edge detection, then voted to get the parameters of iris boundaries by Hough transforms, and the final step used the Fisher Discriminant Linear. However, there is still disadvantages[11]. Wildes method is very computationally demanding because it introduces lots of edge points of other objects, such as eyelashes and eyelids, in Hough transform. The others, which use real-time camera in the research, assume that angle distance of eyes must be straight onto the camera. But, it s quite hard to manage the person to look at the camera with good position that we want. The previous researches also didn t consider that the captured image of iris sometimes closed by eyelid and eyelash. So, it couldn t give a good of iris localization. A. Image Acquisition III. PROPOSED ALGORITHM Image acquisition is considered the most critical step in our research, since all subsequent stages depend highly on the image quality. To get the good quality, it depends on some parameter such as: Lighting Focal Length Aperture Depth of Field Pixel of Resolution After that, the captured image is better to be converted into gray scale image. Images of this sort are composed exclusively of shades of neutral gray, varying from black at the weakest intensity to white at the strongest. B. Pre-Processing This process is to separate the iris from the boundary between iris and pupil. But, it is not simple case. Like Daugman and Wildes method,that used the pre-processing by doing the iris localization in the edge of boundary pupil and iris, didn t get the boundary perfectly. After iris localization, we extract the iris and encode it. Considering that, Pre-processing is divided into two main part: C. Iris Localization 1) Binary: Binary process is the first step in separating the iris and pupil. The idea is to take the pupil as the main point. Since pupil area is dark dot in eye, it is simple to get this area as parameter. So, we assume that pupil area is dark and the others are bright with thresholding. Thresholding changes pixel value to 1 if it s greater than threshold value and 0 in opposite condition. The equation of thresholding is as following: f(i,j) = { 0 if I(i,j) > T 1 if I(i,j) T f(i,j) is the threshold result from condition wether is greater than or less equal then of the threshold value which is T. 2) Dilation: After we got the binary result, we need to reduce the bad effect from threshold value that sometimes narrow the pupil boundary area with dilation[6]. Dilation is one of the basic operations in mathematical morphology. Originally developed for binary images, it has been expanded first to grayscale images, and then to complete lattices. The dilation of A by the structuring element B is defined by: A B = b B A b. The dilation is commutative, also given by: A B = B A = a A B a. If B has a center on the origin, then the dilation of A by B can be understood as the locus of the points covered by B when the center of B moves inside A. The dilation of a square of side 10, centered at the origin, by a disk of radius 2, also centered at the origin, is a square of side 12, with rounded corners, centered at the origin. The radius of the rounded corners is 2. The dilation can also be obtained by: A B = {z E (B s ) z A }, where B s denotes the symmetric of B, and z is the enlargement. 3) Erosion: The dilation result is not enough. We need to concern another condition that the threshold value doesn t narrow the area, but widen it. So, we should widen the pixel with erosion technic. Erosion is one of two fundamental operations (the other being dilation) in morphological image processing from which all other morphological operations are based[6]. The erosion of the binary image A by the structuring element B is defined by: A B = {z E B z A}, where B z is the translation of B by the vector z, i.e., B z = {b + z b B}, z E. When the structuring element B has a center (e.g., B is a disk or a square), and this center is located on the origin of E, then the erosion of A by B can be understood as the locus of points reached by the center of B when B moves inside A. For example, the erosion of a square of side 10, centered at the origin, by a disc of radius 2, also centered at the origin, is a square of side 6 centered at the origin.

3 The erosion of A by B is also given by the expression: A B = b B A b. 4) Deviation: Then, we calculate the value between dilation and erosion. So, we get the deviation of them. It is important to do, since we want to reduce error of pupil boundary area localization. The basic idea in binary morphology is to probe an image with a simple, pre-defined shape, drawing conclusions on how this shape fits or misses the shapes in the image. This simple probe is called structuring element, and is itself a binary image (i.e., a subset of the space or grid)[6]. The equation is defined by: C = {A B} {A B}, where A B = {z E (B s ) z A } is dilation result, and A B = {z E B z A}, is erosion result. 5) Skeleton: The next step is to do the skeleton algorithm. The skeleton algorithm is used to rarefy the edge boundary result from the deviation step. Because, the deviation results a edge boundary that has thickness greater than one pixel. So, we must make the thickness become fit[1]. The advantage of doing this is that it don t get data redundancy since each pixel has data information. The equation is represented by: S(A) = i S i (A) i=0 where S i (A) = (A ib) (A ib) B is elemen structure of its. 6) Pupil Edge Boundary Tracing: This step is to trace the pupil edge boundary. The tracing itself is to separate the pupil edge boundary from other things that sometimes has same pixel like pupil area. In the beginning, we have assume that pupil is a dark area meaning that every dark area on image can be pupil area. What if the condition like eyelashes or other things can be said that it is also a dark area. So, to trace the edge we use Freeman Chain Code[8]. The Freeman chain code is a sequence of directions of the steps taken when following the boundary of a region. Let us define the anticlockwise Freeman code as in figure 3. The inner boundary tracing algorithm can be used to follow the boundary in the image. The algorithm is defined as: 1) Search the image from top left until a pixel P 0 belonging to the region is found. For 4-connectivity assign d = 3, which stores the the direction of the previous move. 2) Search the neighborhood of the current pixel for another pixel P i of the boundary in an anti-clockwise direction beginning from (d + 3)mod4. Update the value of d. 3) If the current boundary element is equal to P 1 and the previous P 0, then stop. Otherwise, goto step 2. 4) The detected inner border is represented by pixels P 0... P n 2. 7) Pupil Center Point Searching: This step is trying to find the pupil middle point. The reason is to determine the pupil area, so we can get the pupil circle area correctly. Then, it can be used in processing the iris feature extraction and (b) (c) (a) FIG. 3. Freeman Algorithm: (a) Step 1 (b) Step 2 (c) Step 3-4 encoding. To do that, we assume that the pupil area is a geometry plane. Since it s an image, the geometry plane is defined into 2D coordinates. One way to do is with Centroid method. Calculating the centroid involves only the geometrical shape of the area. Integration formulas for calculating the Centroid are: C x = xda yda A C y = A A = f(x)dx where the distance from the y axis to the centroid is C x, the distance from the x axis to the centroid is C y, and the coordinates of the centroid are (C x,c y ). D. Feature Extraction and Encoding This part will explain about to extract the feature image and encode it. As assumed before, extracting feature from the image is not simple one. Differ with Daugman s and Wildes feature extraction, we extract the area with two step. First, we do the pupil and iris area cutting. Next step is to encode it with Canny Operator Detection. 1) Iris Area Cutting: The aim of this step is to pick some area of iris and pupil that really represent their feature. So, we don t need all area of iris and pupil since iris area image may be closed by eyelashes or eyelids in real. Depends on the condition, we pick the iris area by cutting it. But, we should care about the height and width of the cutting area. The cutting area can be defined by mathematic equation as following: A = H W where H = {2 H m } is the height of cutting area, and W = {2 W m } is the width of cutting area. H m is the pupil radius minimum area. W m is the iris radius minimum area. Both of H m and W m must be multiplied by 2 since they are radius. Those are pixel value. 2) Encoding: In encoding, the first is edge detection. Edge detection can make the iris circle parameter calculation easier.

Considering the Gaussian function in one dimension, this may be expressed G(x) = 1 2πσ e x2 and the first derivative is G (x) = x e x2 2πσ 3 and the second derivative [ is G (x) = 1 e x2 2πσ 1 x2 3 σ

4 One of the edge detection algorithm is Canny edge detector[3]. The approach was based strongly on convolution of the image function with Gaussian operators and their derivatives, so we shall consider these initially. Considering the Gaussian function in one dimension, this may be expressed G(x) = 1 2πσ e x2 and the first derivative is G (x) = x e x2 2πσ 3 and the second derivative [ is G (x) = 1 e x2 2πσ 1 x2 3 σ 2 ] In fact, the first derivative of the image function convolved with a Gaussian, g(x,y) = D [Gauss(x,y) f(x,y)] is equivalent to the image function convolved with the first derivative of a Gaussian, g(x,y) = D [Gauss(x,y)] f(x,y) Therefore, it is possible to combine the smoothing and detecting stages into a single convolution in one dimension, either convolving with the first derivative of the Gaussian and looking for peaks, or with the second derivative and looking for zero crossings. E. Matching This part is the last step of the improved method. The aim is to match iris meaning the code. The matching process is to get the similarity and dissimilarity value. Differ from Daugman that only use one matching algorithm, we used two algorithm to improve the matching result. We used Hamming and Euclidean Distance Algorithm. 1) Dissimilarity: Hamming Distance For binary strings a and b the Hamming distance is equal to the number of ones in a XOR b[7]. The equation of Hamming Distance is defined by: HD = 1 2,408 2,408 j=1 A j(xor)b j Where A is the first image matrices and B is the second image matrices. Euclidean Distance The Euclidean distance or Euclidean metric is the ordinary distance between two points that one would measure with a ruler, which can be proven by repeated application of the Pythagorean theorem[10]. The formula is defined by: To get the distance value is by calculating the deviation between their coordinates. After that, we square the root of result. Remember that, we calculate the pixel value that located in the same coordinate of its. 2) Similarity: In similarity, the idea is same like dissimilarity. We calculate the distance from two image that we want to match. However, we can use only HD or Euclidean to match it. But, we want to know more about the distance specially in similarity. Consider to the basic idea in matching images, we should calculate its pixel. Mathematically, the formula can be written as following: n i=1 S i = P i + ((C h,i )x(i h,i )) n i=1 ((Q i + (I h,i ) 2 )x(q i + (C h,i ) 2 ) IV. TESTING AND RESULT In this part, we are discussing about testing system. We have used two different cameras to acquire the image meaning the input. First camera is a simple web camera with specification: 240 x 320 CMOS Resolution,1.3 Megapixel Camera, USB 2.0. And the second is Casia s camera with specification: Unknown (Made by Pattek Corporation and 280 x 320 CCD Resolution). We have also processed the system using Matlab version 7.0. In our research, first we have tested 4 condition way with two cameras. FIG. 4. Input Image:(a) Room Lighting, (b) Side Lighting (c) With Eyeglasses (d) Normal Lighting As seen in figure 4, image (a), (b), and (c) was taken by simple web camera, and image (d) was taken by Casia s camera. The result of iris localization seen in figure 5. FIG. 5. Iris Localization After doing the process, we got each iris code seen in figure 6. p1(x 1,y 1 )andp2(x 2,y 2 ) = ((x 1 x 2 ) 2 + (y 1 y 2 ) 2 ) where p1(x 1,y 1 ) is first image with x and y coordinates of pixel. where p2(x 2,y 2 ) is second image with x and y coordinates of pixel. FIG. 6. Iris Code

Total Time Curve the shiny lighting can influenced the image pixel. Because, the shiny image changes the pixel from the low intensity to high intensity.

5 During the our research, we have done the experiment and test using 30 data. Finally, we got the extract time meaning the time needed for obtaining the feature extraction and total time for whole computing taken as seen in figure 7. FIG. 9. Total Time Curve the shiny lighting can influenced the image pixel. Because, the shiny image changes the pixel from the low intensity to high intensity. As higher the intensity, the pixel becomes higher too. the glasses or mirror can spare the reflection of light from the environment. So, the captured image is difficult to find the iris localization FIG. 7. Time Computing Table As seen in figure 8 and 9, showing the differences between the data taken by simple web camera and Casia s camera for 30 data. FIG. 8. Extract Time Curve V. CONCLUSION In conclusion, Our proposed method did quite good in solving the problem. There are some points that we could underlined as following: the resolution of camera can influenced the depth of field in image captured. The resolution in here depends on focal length of camera and its aperture. But, when the input is good image like from Casia, the algorithm is powerful to get the iris localization. The boundary edge is fit well. According to table and curves, the time processing between simple web camera and Casia s camera is not quite different. However, this algorithm still has to be improved in matching proses. Because, the value of iris code depends on the pixel value. Eventhough, the images are from the same person, the similarity value is not closer to value 1. So, we should use the camera that have good quality. Therefore, it ll be nice if we can implement or do the acquisition only with simple camera in future work. REFERENCES [1] A. S. M. Bitter, I.; Kaufman. Penalized-distance volumetric skeleton algorithm. volume 7, pages IEEE Transactions, September Visualization and Computer Graphics. [2] B. M.-I. T. Burghardt. Inside iris recognition. Master s thesis, University of Bristol, November [3] J. Canny. A computational approach to edge detection. pages IEEE Trans Pattern Analysis and Machine Intelligence, November [4] J. Daugman. The importance of being random: Statistical principles of iris recognition. 36(2): , [5] J. Daugman. New methods in iris recognition. IEEE Transactions on Systems, Man, and Cybernetics, [6] E. R. Dougherty. An Introduction to Morphological Image Processing. Bellingham, Wash, USA, SPIE Optical Engineering Press. [7] R. W. Hamming. Error detecting and error correcting codes. Bell System Technical, pages , [8] J. S. Jukka Iivarinen, Markus Peura and A. Visa. Comparison of Combined Shape Descriptors for Irregular Objects, [9] M. Nabti and B. Ahmed. An Improved Iris Recognition System Using Feature Extraction Based on Wavelet Maxima Moment Invariants, volume Springer Berlin / Heidelberg, August [10] T. Saito and J. Toriwaki. New algorithms for euclidean distance transformations of an n-dimensional digitized picture with applications. 27: , Pattern Recognition. [11] R. Wildes. Iris recognition: an emerging biometric technology. volume 85 of i9. Proc. IEEE.

Tutorial 8. Jun Xu, Teaching Asistant March 30, COMP4134 Biometrics Authentication

Tutorial 8. Jun Xu, Teaching Asistant March 30, COMP4134 Biometrics Authentication Tutorial 8 Jun Xu, Teaching Asistant csjunxu@comp.polyu.edu.hk COMP4134 Biometrics Authentication March 30, 2017 Table of Contents Problems Problem 1: Answer The Questions Problem 2: Daugman s Method Problem