A Fast Circular Edge Detector for the Iris Region Segmentation Yeunggyu Park, Hoonju Yun, Myongseop Song, and Jaihie Kim I.V. Lab. Dept. of Electrical and Computer Engineering, Yonsei University, 134Shinchon-dong, Seodaemun-gu, 120-749 Seoul, Korea parkyk@seraph.yonsei.ac.kr http://cherup.yonsei.ac.kr Abstract. In this paper, we propose a fast circular edge detector for the iris region segmentation by detecting the inner and outer boundaries of the iris. In previous work, the circular edge detector which John G.Daugman proposed, searches the radius and the center of the iris to detect its outer boundary over an eye image. To do so, he used Gaussian filter to smooth texture patterns of the iris which cause its outer boundary to be detected incorrectly. Gaussian filtering requires much computation, especially when the filter size increases, so it takes much time to segment the iris region. In our algorithm, we could avoid procedure for Gaussian filtering by searching the radius and the center position of the iris from a position being independent of its texture patterns. In experimental results, the proposed algorithm is compared with the previous ones, the circular edge detector with Gaussian filter and the Sobel edge detector for the eye images having different pupil and iris center positions. 1 Introduction As the importance of the personal identification increases, the researches on the biometric technologies which use person s unique physical features are actively performed. The major biometric technologies are iris recognition, fingerprint verification, face recognition, and so on [1]. The iris recognition guarantees higher security than any other biometric, since the iris has highly detailed and unique textures which remain unchanged [2][3][4]. The iris recognition system verifies and identifies a person using these characteristics. In general, the texture images have many small edge components representing local information. In the iris region segmentation, there may be obstacles because of them, so they should be removed to detect the correct inner and outer boundaries of the iris. So, John G.Daugman proposed the circular edge detector with Gaussian filter. However, it requires heavy computational complexity because of using Gaussian filter for smoothing texture patterns of the iris which cause its outer boundary to be detected incorrectly [2][5][6]. So we propose a fast circular edge detector in which Gaussian filtering is not necessary by searching the radius and the center position of the iris from a position being independent of the texture patterns of the iris. S.-W. Lee, H.H. Bülthoff, T. Poggio (Eds.): BMCV 2000, LNCS 1811, pp. 417 423, 2000. c Springer-Verlag Berlin Heidelberg 2000
418 Y. Park et al. Fig. 1. Iris recognition procedure 2 Iris Recognition Procedure The general iris recognition procedure is shown in Fig. 1. In the above procedure, it is important to segment the iris region, exactly and fast. After all, it affects the generation of the iris code and the recognition performance. 3 Iris Region Segmentation The iris region segmentation is locating the iris region from an eye image like Fig. 2. At first, we should detect the inner boundary between the iris and the pupil and then outer one between the iris and the sclera. In general, an iris has not only the texture patterns including its local characteristics but also many edge Fig. 2. Eye image
A Fast Circular Edge Detector for the Iris Region Segmentation 419 components inside it as shown in Fig.2. More features affecting the performance of the iris region detection exist close to the inner region than the outer one [2], especially, in case of brown color iris. Therefore, the typical edge detection operators such as Sobel or Prewitt operator may not be good to segment the iris region. 3.1 Circular Edge Detector with Gaussian Filtering John G.Daugman proposed the circular edge detector with Gaussian filter to segment the iris region. The role of Gaussian filtering is to smooth the texture patterns inside it to avoid detecting the outer boundary of the iris, incorrectly such as shown in Fig. 3 [2]. However, Gaussian filtering requires much computa- Fig. 3. Example of locating incorrect outer boundary of the iris tion time for segmenting the iris region, especially when filter size increases. His algorithm searches the radius and the center position of the iris for detecting its outer boundary over an eye image by using Eq. (1). That requires much computation time, too. This method was implemented by integrodifferential operators that search over the image domain (x, y) for the maximum in the blurred partial derivative, with respect to increasing radius r of the normalized contour integral of I(x, y) along a circular arc ds of radius r and center coordinates (x 0,y 0 ): max(r, x 0,y 0 ) G σ I(x, y) r r,x 0,y 0 2πr ds (1) where * denotes convolution and G σ (r) is smoothing function such as a Gaussian of scale σ [2][3][4][5][6][7]. 3.2 Sobel Edge Operator The Sobel operator is one of the most commonly used edge detectors [8][9][10][11]. In order to evaluate our algorithm, we applied the 8-directional Sobel operator to segment the iris region.
420 Y. Park et al. 3.3 The Proposed Fast Circular Edge Detector This method consists of five steps to segment the iris region as shown in Fig 4. Fig. 4. Segmentation steps At first, we detect a pupil region to locate its center position. Detecting the pupil region is achieved by thresholding an eye image. In step 2, we locate the pupil center position based on the histogram and long axis of the pupil. In step 3, we can easily find the iris inner boundary by using Eq. (2). 1 max(n r, x 0,y 0 ) I(x, y) (2) r k m x = k r cos(m θ + x 0 ) y = k r sin(m θ + x 0 ) where r and θ are radius interval and angular interval, respectively. In step 4 and step 5, we locate each iris center and its outer boundary. In these steps, Gaussian filtering may be necessary, since the iris region has the texture patterns similar to random noise. If we carefully analyze the pattern of the iris region, we could cognize it to be the smoother, the more outward. Using this trait, we need to start at radius r being independent of its texture patterns
A Fast Circular Edge Detector for the Iris Region Segmentation 421 to detect the iris outer boundary. The start position r s is computed from Eq. (3). x =2r i (3) r s =2r p If x =2r p =2r i + d (= r p r i ) otherwise where r i, r p and x are the ideal pupil radius, the real pupil radius and the Fig. 5. Ideal pupil radius and reference point reference point being independent of the texture patterns of the iris, respectively. Therefore, in our algorithm, Gaussian filtering requiring much computation time is not necessary. 4 Experimental Results To evaluate the performance of algorithms for the iris region segmentation, we experiment in the following environment and for the eye images having the different pupil and iris center positions. * Input image size 640 X 480 pixels gray level image. * Camera focus : fixed * Center of iris and pupil : not concentric * Camera : black and white CCD camera * CPU : Pentium II Xeon 450MHz We could acquire the result images using the iris segmentation algorithms as shown in Fig. 6. As you can see in Fig. 6, the accuracy of the proposed algorithm is nearly equal to Daugman s one, but Sobel edge operator could not locate the correct iris region. It is why the edge of the outer boundary of the iris faints, and the iris region has many edge components. In processing time, our algorithm is superior to any other method like Table 1. As you can see from the experimental results, because Daugman s algorithm
422 Y. Park et al. Fig. 6. Input and result images uses Gaussian filter and searches the radius and the center position of the iris for detecting its outer boundary over an eye image, his algorithm takes much processing time than ours in detecting the iris region. 5Conclusions In this paper, we proposed a fast circular edge detector to segment the iris region from a person s eye image. In previous work, Daugman proposed circular edge detector with Gaussisn filter. His method has some factors increasing the computational complexity. We proposed a fast algorithm which Gaussian filter-
A Fast Circular Edge Detector for the Iris Region Segmentation 423 Table 1. Processing time(sec) ing was not necessary by using reference point being independent of the texture patterns of the iris. As results, our algorithm is faster than Daugman s one. References 1. Despina Polemi.: Biometric Techniques: Review and Evaluation of Biometric Techniques for Identification and Authentication, Including an Appraisal of the Areas where They Are Most Application, Institute of Communication and Computer Systems National Technical University of Athens (1999) 5-7, 24-33 2. J. G. Daugman.: High Confidence Visual Recognition of persons by a Test of Statistical Independence, IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol. 15, NO. 11 (1993) 1148-1160 3. J. G. Daugman.: Iris Recognition for Persons Identification (1997) 4. R. P. Wildes.: Iris Recognition-An Emerging Biometric Technology, Proceedings of the IEEE, Vol. 85, NO. 9 (1997) 1348-1363 5. N. Chacko, C. Mysen, and R. Singhal.: A Study in Iris Recognition (1999) 1-19 6. J. G. Daugman.: Recognizing Persons by Their Iris Patterns, Cambridge University (1997) 1-19 7. D. McMordie.: Texture Analysis of the Human Iris, McGill University (1997) 8. R. Jain, R. Kasturi, and B. G. Schunk.: Machine Vision, McGrow-Hill (1995) 145-153 9. E. Gose, R. Johnsonbaugh, and S. Jost.: Pattern Recognition and Image Analysis, Prentice Hall PRT, Inc (1996) 298-303 10. M. Nadler and E. P. Smith.: Pattern Recognition Engineering, Jonh Wiles & Sons, Inc (1993) 107-142 11. M. Sonka, V. Hlavac, and R. Boyle.: Image Processing, Analysis, and Machine Vision, Brooks/Cole Publishing Company (1999) 77-83