100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 Abstract Iris recognition is increasingly employed as a biometric modality to improve the security across the government, public, and private sectors; however, the limitations of iris recognition are yet to be fully understood. Several approaches have been previously proposed that allow performing a reliable identification of a person based on iris information. However, resulting iris identification systems are limited in their ability to correctly identify a person because of the imperfectly acquired iris images. Two iris image properties resolution and blur - significantly impact the identification performance. The impact of these two iris image properties on correct recognition rates have been explored and summarized in this work in an attempt to provide clarity for challenges related to iris biometrics. 1. Introduction As the level of security breaches and transaction fraud increases, the need for highly secure identification and personal verification technologies is becoming apparent. This need for biometrics can be found in federal, state, and local governments, in the military, and in commercial applications. It is clear that every entity which provides access to private information can greatly benefit from a secure identification and personal verification system. Many systems have been proposed as a solution for such a need including but not limited to fingerprinting, palm printing, vascular pattern recognition, hand geometry, dynamic signature, voice recognition, facial recognition, and iris identification [1]. 1.1. Motivation of Iris Identification Impact of Resolution and Blur on Iris Identification Clark Phillips (cp1038@txstate.edu), Oleg V. Komogortsev (ok11@txstate.edu) In order for a human characteristic to be used as a biometric, it must meet specific criteria. Any potential biometric characteristic is analyzed against the biometric rubric. This rubric consists of qualifiers: universality (each person should have the characteristic), uniqueness (how well the characteristic separates individuals), permanence (how well the characteristic resists aging and other time dependent variances), collectability (ease of acquisition for measurement), performance (accuracy, speed, and robustness of the technology used), acceptability (degree of approval of a technology), and circumvention (ease of use of a substitute for the characteristic) [2]. Irises are perceived to be one of the best characteristics to use as a biometric trait because the error rate of iris recognition is one of the lowest among known biometric traits [3]. Majority of people have two irises from birth until death. No two irises are considered to be the same [3]. Irises can contain many distinctive features such as arching ligaments, furrows, ridges, crypts, coronas, freckles, and zigzag collarets [4]. The iris has the great mathematical advantage that its pattern variability among different persons is enormous [5]. Irises remain stable from six months of age until death [6]. Since the iris is an internal organ that can be seen externally, iris identification systems can be noninvasive [5]. Burch's initial proposal of using the iris as a pattern for identification has become a reality [6]. The iris pattern is considered to be difficult to reproduce [7], however even commercial iris recognition systems have frequently accepted reproduced irises as legitimate irises [8]. Regardless of the contemporary research on iris recognition, important questions regarding robustness of iris identification in various usability scenarios remain. Among those there are questions concerning the limitations of iris identification for cases of extreme degradation of image quality. The objective of this paper is to provide a better understanding of the impact of captured iris dimensions and image blur on corresponding identification accuracy. 1.2. Related Work Many solutions have been proposed for answering the demand for an iris identification system. Improvements have been proposed to these systems in an attempt to optimize their performance. However, precious little is written concerning some of the limitations associated with all iris identification systems, namely image resolution and image blur. Although the non invasive nature and high variability of irises make them a promising biometric, high quality photographs with high resolution and good contrast are needed for today s iris identifying algorithms to perform 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 1
200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 well. This is not a problem if the iris is photographed at very close range with a cooperative subject, but if the subject is farther away and possibly even walking, photographing the iris for identification becomes more challenging [9]. According to Daugman, a minimum radius of 70 pixels (140 pixel diameter) is required in order to capture the rich details of the iris pattern [10]. However, ISO/IEC has set the standard of a required minimum 200 pixel diameter across the iris for high quality iris images [10]. Hollingsworth's research in how dilation of the pupil affects biometric performance suggests that the diameter of the iris is an inadequate way to measure the total amount of iris available. She asserts that the annular width of the iris is a more correct measure of the iris size [11]. Algorithms have been proposed to de-blur images [12]. These algorithms are employed when an image is found to be sufficiently blurry, cannot be re-acquired, and whose correct iris identification is less than sufficient. Some proposals are designed to select the best quality iris image from a set of images [13]. Studying the effect of severe image compression on iris recognition performance has been pursued [14]. The compressed images may be perceived to be blurred although there is a significant difference between a blurred image and a compressed image. To the best of our knowledge, there is no published work that investigates in detail the relationship between image blur and correct iris recognition. 2. Objectives The objective of this research is to investigate the relationship between: 1) Image Resolution and Identification Performance and 2) Image Blur and Identification Performance. 3. Methodology Four components are required to test the effect of the iris size and image blur on iris recognition rates: 1) An iris database 2) Correct recognition rate definition 3) The ability to alter the image resolution 4) The ability to alter image blur 5) Iris identification software. 3.1. Iris Database The UBIRIS.v1 database [15] was selected among other databases (CASIA-IrisV3-Interval [21], MMU iris [22], etc.) because it provides high quality iris images with minimized noise factors and is widely accepted among iris recognition researchers. 3.2. Correct Recognition Rate Definition Correct Recognition Rate (CRR) was defined as a true accept rate [20] of a biometric system. 3.3. Image Resolution Our goal was to test iris image resolution requirements. In order to define the effect of image resolution on iris recognition, several datasets of varying iris image resolutions were created. The initial dataset was simply the raw first session recorded UBIRIS.v1 database. This dataset was then copied twelve times. The resolution of each copy's iris images were then decreased. Microsoft Office Picture Manager was used to alter image resolution. Microsoft Office Picture Manager allows the resolution of multiple images in multiple folders to be changed simultaneously. A C The above iris resolutions were created starting with a very small resolution and reaching maximum available size. Table 1 presents tested resolution levels. Iris Annular Width Sample Iris Diameter B Figure 1 : A) 1% resolution B) 50% resolution C) 100% resolution Image Width Image Height Pixels 3 8 20 15 300 7 20 45 34 1530 10 28 63 47 2961 14 40 89 67 5963 17 49 110 82 9020 19 56 126 95 11970 22 63 141 106 14946 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 2
300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 24 69 155 116 17980 26 75 167 125 20875 28 80 179 134 23986 29 85 190 142 26980 31 90 200 150 30000 Table 1: Iris Image Resolution Information 3.4. Image Blur Our goal was to blur the image in the same way as done by an out of focus camera. In photography, bokeh is the blur or aesthetic quality of the blur [16, 17]. Bokeh can be simulated by convolving the image with a kernel that corresponds to the image of an out-of-focus point source taken with a real camera. Gaussian blur is less computationally expensive and produces a softer effect than convolution. The Gaussian blur filter uses a Gaussian function for computing the alteration applied to each pixel. The equations of a Gaussian function in one and two dimensions: Equation 1: 1-Dimensional ( ) Equation 2: 2-Dimensional ( ) In order to apply a Gaussian function to an image, a C++ program was developed that can apply a Gaussian blur to multiple images in multiple directories. Librow's Gaussian filter implementation was used [18]. This filter is based on a window. As the window size increases, so does the blur. To produce the largest possible blur for our database, we searched for the lowest window size corresponding to a CRR of 0%. We found that the lowest window size which produced a CRR of 0% was 85 (σ=85). The window size of 0 represents 0% blur which corresponds to the original dataset. Twenty-six unique window sizes were selected between those extreme points to represent the whole range of possible blur levels. A dataset was created for each of the 26 selected window sizes. Table 2 presents tested blur levels. Window Size (σ) % Blur 0 0.00% 1 1.18% 3 3.53% 5 5.88% 7 8.24% 9 10.59% 11 12.94% 13 15.29% 15 17.65% 17 20.00% 21 24.71% 25 29.41% 33 38.82% 39 45.88% 43 50.59% 47 55.29% 51 60.00% 59 69.41% 61 71.76% 65 76.47% 67 78.82% 71 83.53% 77 90.59% 81 95.29% 83 97.65% 85 100.00% Table 2: Iris Image Blur Information A B Figure 2: A) Window size 9 B) Window size 85 3.5. Iris Recognition Software The software suite that was selected was GIRIST, a free iris recognition system [19]. GIRIST performs comparably to the commercial systems deployed today [23]. GIRIST is able to compute CCR to assess accuracy of identification. 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 3
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 GIRIST produces 96.8% CRR on average when tested on the following databases: CASIA-IrisV3-Interval [21], UBIRIS.v1 [15], MMU iris [22]. 4. Results 4.1. Resolution Impact Figure 3 presents CRR results for the iris resolutions represented by Table 1. Figure 3: Correct Recognition Rate as a function of Iris Diameter The CRR decreases sharply when iris diameter approximately becomes 70 pixels (Figure 3). The next drop off point occurs at sample iris diameter 49. These saturation points are important because increasing beyond the resolution suggested by those points does not seem to provide a substantial increase in identification accuracy. 4.2. Resolution Impact Figure 4 presents CRR results for the iris blur represented by Table 2. Figure 4: Correct Recognition Rate as a function of Gaussian Blur Window Size Figure 4 shows that as blur increases, CRR decreases. CRR only drops ~1.5% as blur increases from 0% to 5%. The recognition rate decreases linearly as blur increases from 10% to 30% (~10% blur ~90% CRR, ~20% blur ~80% CRR, ~30% blur ~70% CRR). A big drop occurs near 30% blur. Unfortunately, this drop is not extremely significant because the CRR has already decreased substantially and therefore the behavior past this point is trivial. This graph demonstrates very clearly that it is possible for researchers to sacrifice a small amount of blur on the entire image in an attempt to reduce the image's high frequency components. 5. Limitations 5.1. Image Blur The Gaussian blur is an emulation of depth of field blur found in common photographs, and thus the inferences we may make from this research may not pertain to all blur situations. Therefore further research is necessary to define the relationship between motion blur and correct iris identification. 5.2. Iris Recognition Software GIRIST was selected largely because of its performance and its availability. While it is not open source, the executable is free. No other iris recognition software was used to corroborate our results. Other iris recognition software may produce different results. 5.3. Image Database A single database was used in this research [15]. Therefore the limits are limited to this single database. For example, the diameters of the irises in these images are around 90 pixels. We are limited by this maximum resolution. The behavior of CRR for iris diameters above 90 pixels is unknown. Another example is that the images contained within this database are not color. The effect of color images is unknown. 6. Conclusion & Future Work Iris recognition is increasingly employed as a biometric modality to improve the security across the government, public, and private sectors; however the limitations of iris recognition are yet to be fully understood. This work explored the impact of image resolution (with corresponding iris width) and image blur on correct recognition rates. 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 4
500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 Results received as a part of the iris resolution investigation indicate the iris width may be significantly smaller than what was previously perceived to provide acceptable levels of recognition. The image resolution analysis saturation points provide evidence that resolution may be sacrificed at the expense of perhaps a more economical camera, a more portable camera, a more convenient camera such as a webcam, or data storage. Though saturation points are apparent in the preceding data, it is uncertain whether there are further saturation points at increasing resolution levels. This necessitates further research into the relationship between image resolution and correct iris identification. Quantitative analysis indicates that a small amount of blur can be employed to reduce potential image noise without losing a significant amount of identification accuracy. We feel that the presented results are useful for the practitioners that use iris-identification systems because they allow better understanding of the limitations and capabilities or such systems. Additional research should be conducted to understand the impact of different types of blur including motion blur, resolution degradation on a wider range of iris recognition techniques, and image databases. 7. Acknowledgements This word was partially supported by grant #60NANB10D213 from the National Institute of Standards (NIST). References [1] B. L. Jay R Bhatnagar, RK Patney, "Performance issues in biometric authentication based on information theoretic concepts: A review," IEEE Technical Reviews, vol. 27, pp. 273-285, 2010. [2] A. K. R. Jain, Arun; Prabhakar, Salil, "An introduction to biometric recognition," IEEE Transactions on Circuits and Systems for Video Technology, vol. 14, pp. 4-20, 2004. [3] L. F. a. A. Safir, "Iris Recognition System," US Patent 4 641 649, 1987, 1987. [4] J. Daugman, "The Importance of Being Random: Statistical Principles of Iris Recognition," Pattern Recognition, vol. 36, 2003. [5] T. T. L. Ma, Y. Wang, and D. Zhang, "Efficient Iris Recognition by Characterizing Key Local Variations," IEEE Transactions on Image Processing, vol. 13, 2004. [6] Iris Recognition. Available: http://www.biometrics.gov/documents/irisrec.pdf [7] R. K. G. Annapoorani, P. Gifty Jeya, and S. Petchiammal, "Accurate and Fast Iris Segmentation," International Journal of Engineering Science and Technology, vol. 2, 2010. [8] J. Daugman, "Iris Recognition and Anti-Spoofing Countermeasures," The 7th International Biometrics Conference, 2004. [9] U. K. A. Nadeau, P. Lopez-Meyer, and S. Schuckers, "Iris Identification from a Distance," Honors Thesis, Clarkson University Department of Electrical and Computer Engineering, Clarkson University. [10] I. O. f. Standardization, "Information Technology - Biometric data interechange formats " in Part 6: Iris image data vol. ISO/IEC 19794-6:2005, ed: International Organization for Standardization, 2005, p. 25. [11] K. B. K. Hollingsworth, and P. Flynn, "Pupil Dilation Degrades Iris Biometric Performance," Computer Vision and Image Understanding, vol. 113, 2009. [12] L. R. X. Huang, and R Yang, "Image Deblurring for Less Intrusive Iris Capture," IEEE Conference on Computer Vision and Pattern Recognition, 2009. [13] R. M. Y. Lee, and P. Phillips, "Robust Iris Recognition Baseline for the Ocular Challenge," IEEE Conference on Automatic Face and Gesture Recognition, 2011. [14] J. D. a. C. Downing, "Effect of Severe Image Compression on Iris Recognition Performance," IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, vol. 3, pp. 52-61, 2008. [15] UBIRIS.v1 [Online]. Available: http://iris.di.ubi.pt/ubiris1.html [16] H. Davis, Practical Artistry: Light & Exposure for Digital Photographers: O'Reilly Media, 2008. [17] T. Ang, Dictionary of Photography and Digital Imaging: The Essential Reference for the Modern Photographer: Watson Guptill, 2002. [18] S. Chernenko. Gaussian Filter, or Gaussian Blur. Available: http://www.librow.com/articles/article-9 [19] "GRUS IRIS TOOL ", ed. [20] P. G. R. M. a. P. J. Phillips, "Face Recognition Vendor Test 2002 Performance Metrics," Fourth International Conference on Audio-Visual Based Person Authentication, 2003. [21] CASIA-IrisV3 [Online]. Available: http://www.cbsr.ia.ac.cn/english/irisdatabase.asp [22] MMU1 Iris [Online]. Available: http://pesona.mmu.edu.my/~ccteo/ [23] GIRIST Tutorial. Available: http://www.grusoft.com/girist/girist_tutorial.pdf 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 5