All Iris Code Bits are Not Created Equal

Transcription

1 All ris Code Bits are Not Created Equal Karen Hollingsworth, Kevin W. Bowyer, Patrick J. Flynn Abstract-Many iris recognition systems use filters to extract information about the texture of an iris image. n the Daugman-style approach, the filter output is mapped to a binary iris code. The normalized Hamming distance between two iris codes is computed and decisions about the identity of a person are based on the computed distance. The normalized Hamming distance weights all bits in an iris code equally. However, this work presents experimental evidence that all the bits in an iris code are not equally useful. Some bits are more consistent than others. An explanation of the cause of this discrepancy is given. Different regions of the iris are compared to evaluate their relative consistency. Finally, this paper investigates the theoretical impact of consistent and inconsistent bits on the false reject rate of an iris recognition system.. NTRODUCTON THE human iris has extraordinary variations in texture, and it has been shown that iris texture can be used to identify a person. One typical method of using the iris for recognition involves applying a texture filter to an image of the iris, and extracting a representation of the texture, called the iris code. The iris code is a set of bits, each one of which indicates whether a given texture filter applied at a given point on the iris image has a negative or non-negative result. Many different texture filters and many different lengths of iris code have been investigated, although the Gabor filter texture analysis proposed by Daugman [1] is considered the traditional approach. Readers interested in other approaches may look at [2] and [3]. n addition, [4] presents a comparative study of different iris recognition algorithms. For a given iris, a bit in its iris code is "fragile" if there is any substantial probability of it ending up a 0 for some images of the iris and a 1 for other images of the iris. The possibility that fragile bits exist in the iris code was initially suggested by Bolle et al. [5]. They model the theoretical lower bounds for the false accept rate (FAR) and the false reject rate (FRR) of an iris biometrics system, and find that "the reported empirical FRR performance degradation is significantly more stable with respect to the system threshold variation than predicted by the theory. This implies that the invariant bits in the iris code representation are dramatically robust to the imaging noise." As a possible explanation for the difference between the empirical and theoretical results, they suggested "...that perhaps not all bits are equally likely to flip, that there are some particularly 'fragile' bits." n this paper, we use the terms "fragile" and "inconsistent" Karen Hollingsworth, Kevin Bowyer, and Patrick Flynn are with the University of Notre Dame. Telephone: (574) Fax: (574) addresses: kholling ond.edu, kwb 0cse.nd.edu, flynn o,nd.edu. interchangeably. To investigate the existence of fragile bits, we compared iris codes from multiple images of the same eye. This comparison showed that, indeed, some bits are more reliable than other bits in an iris code.. DATA AND SOFTWARE The data used for this experiment was collected at the University of Notre Dame and provided to the National nstitute of Standards and Technology (NST) for use in the ris Challenge Evaluation (CE) [6]. All images were acquired using an LG2200 camera. This dataset contains over a hundred different images for each of several irises. No more than six images were acquired for a particular iris of a particular subject in any given week. To test the fragility of individual bits in an iris code, we selected images that were mostly unoccluded by eyelids or lashes. Examples images are shown in Figures 1 through 4. We selected a subset ofthe data that contained 24 subjects, with between 15 and 118 images of the left eye of each subject, for a total dataset of 1251 images. This dataset contains 4 Asian and 20 Caucasian subjects. Ten of the subjects are female and 14 are male. Fig 1. Example image from our dataset. This image has no occlusion by eyelids or eyelashes. However, part of the sclera is included in the segmentation of this iris image. We used software similar to the open source software, risbee [6], for creating the iris codes. This software uses one dimensional log-gabor wavelets to create a 240x20x2-bit iris code. The software we used contains improvements to the segmentation as described in [7]. Our software found the inner and outer boundaries of the iris correctly, as determined by visual inspection, in all 1251 images selected for the study. The images were selected to /07/$ EEE Authorized licensed use limited to: UNVERSTY NOTRE DAME. Downloaded on December 21, 2009 at 15:14 from EEE Xplore. Restrictions apply.

2 there are specular highlights on irises of some of the images (Figure 2). To deal with these segmentation difficulties, we later consider how masking parts ofthe iris affects our results. n cases where the segmentation software detected occlusion by eyelids, parts of the iris region were masked (Figure 3). f part of the iris code was masked in even one image of a subject, the bits from that part of the iris code were left out of all computations in our experiments on that subject. Some of the images have sharp focus, but a few images are poorly focused (Figure 4). Our experiments show that some bits are consistent across all images, even when a few poorly-focused images are included in the experiment. This result matches Bolle's assertion that "the invariant bits in the iris code representation are dramatically robust to the imaging noise." [5] Two different iris codes from the same eye are not necessarily in the same orientation, and may have to be rotated in order to be aligned correctly. We wrote code that would rotate the iris codes so that all codes from the same eye could be aligned and compared for consistency. The output was a 240x20x2 matrix, where each entry in the matrix represented the average value for a particular bit in the iris code. Half of these entries correspond to the real part of the output from the texture filtering process, and half of the entries correspond to the imaginary part. n order to see patterns in the output, we divided each matrix into two 240x20xl matrices where one matrix represents the real bits of the iris code, and the other represents the imaginary bits. Fig 3. n this image, parts of the top and bottom of the iris were masked before the iris code was computed. Fig 4. Despite including some poorly focused images in the dataset, on average 15.99% of bits in the iris codes were perfectly consistent. minimize the effects of segmentation errors; however, some segmentation difficulties could not be avoided. One difficultly lies in the fact that irises are not exactly circular. Therefore, even the best circle caught part of the sclera or eyelid above and below some of the irises (Figure 1). Also,. ExiSTENCE OF FRAGLE BTS All subjects had three different regions apparent in their iris codes: areas consistently equal to 0, areas consistently equal to 1, and inconsistent areas. The inconsistent areas tend to occur at the boundaries between regions of zeros and regions of ones. Examples ofthe inconsistent regions in the iris codes are shown in Figure 5. f a bit was equal to one the majority of the time, but was equal to zero 30% of time, then we say that the bit "flipped" in 300% ofthe iris codes. f a bit was zero the majority of the time, but one for 30% of the time, we still say that the bit "flipped" in 30% of the iris codes. n this figure, the black regions correspond to bits that were flipped in at least 30% of the iris codes. f we are less strict in our definition of what constitutes a truly inconsistent bit, then there will be more inconsistent bits. Figure 6 shows that ifwe consider any bit that flips more than 20% of the time to be fragile, then there will be many fragile bits. fwe consider any bit that flips more than 40% of the time, then there will be fewer fragile bits. n our study we found that, on average, 15.99% of the bits in an iris code were perfectly consistent; that is, 15.99% ofthe unmasked bits were always equal to 1 or always equal to 0, for all iris codes for an iris. The subject with the smallest fraction had 4.74% of the bits perfectly consistent, and the subject with the largest fraction had 33.9% of the bits perfectly consistent. The sample standard deviation across the 24 irises was 6.39%. Authorized licensed use limited to: UNVERSTY NOTRE DAME. Downloaded on December 21, 2009 at 15:14 from EEE Xplore. Restrictions apply.

3 [~C 4~4f45F (Zr -L1C. J4¾J Y Wi F A bea 6 Fig 5. Black areas in each rectangle are less consistent parts of the iris code, and white areas are more consistent. Each rectangle represents data from the iris codes of at least 15 different images of the same eye. Two rectangles are shown for each subject; one rectangle shows the real bits in the iris code, and the next rectangle shows the imaginary bits. Light gray regions are masked regions. The numbers on the side are the subject numbers associated with the CE dataset. (A) p=20% acquire more images of an iris, there are more opportunities to acquire outliers in any given bit in the iris code. n addition to considering perfectly consistent bits, we also looked at the average fragility of the unmasked bits of a subject. For each bit, the percent of the time that the bit flips must lie between 0n and 500n, or between 0 and 0.5. We found the percent ofthe time that each bit flipped and took the average across all bits for a subject. This average fragility for each subject is not correlated with the number of images we had for the subject. The correlation coefficient between these factors is only , which is not statistically significant (p-value ). Fragile bits show up in about equal amounts in both males and females. Figure 7 shows the average fragility of the 24 subjects when divided into groups of males and females. There is no apparent difference between the two genders. n our dataset, Asian subjects had more fragile bits than Caucasian subjects. Figure 8 shows average fragility versus race, and the Asian subjects in general had more fragile bits in their iris codes. However, we only have four Asian subjects in our dataset, so it is quite possible that this trend may not be apparent if we had more Asian subjects. 0, c.a.2._- 0 0, LV F 0.2 F 0.15 F 0.1k 0.05s J Male Female Fig 7. There is no difference in the average fragility of iris code bits between the two different genders. (B) p=30% (C) p=40% Fig 6. The number of fragile bits depends on the threshold used. There are fewer bits that flip in at least 40% of the iris codes than there are that flip in at least 20% of the iris codes. There is a negative correlation between the number of images we had for a subject and the percent of perfectly consistent bits. Specifically, the correlation coefficient between number of images and percent ofperfectly consistent bits is , which is a statistically significant correlation (p-value ). This result makes sense because as we a.2_._- _._ n} k (a) L- 0>u Q.1F _Q r) 0.3k 0.25k 0.2k 0.05 F nj Asian White Fig 8. The four Asian subjects in our dataset tended to have more fragile bits in their iris codes, but with only four subjects, we cannot conclude that this is a general trend. Authorized licensed use limited to: UNVERSTY NOTRE DAME. Downloaded on December 21, 2009 at 15:14 from EEE Xplore. Restrictions apply.

4 The spatial pattern apparent in these consistent and inconsistent regions likely comes as a result of how the iris image is processed. A Fourier transform is applied to the iris image, and then the values are multiplied by the log-gabor filter. Next an inverse Fourier transform is applied, resulting in 4800 complex numbers. Rather than storing the complex numbers as the iris codes, each complex number is mapped to a two-bit number. f the real part of the number is positive, then the first bit is a 1, otherwise the first bit is a 0. f the imaginary part ofthe number is positive, then the second bit is a 1, and otherwise, the second bit is a 0. Thus, a complex number in the first quadrant of the complex plane is mapped to the value, 1 1; the second quadrant, 01; the third quadrant, 00; and the fourth quadrant, 10. f a region ofthe iris image were associated with a complex number near the negative y-axis, small changes in the complex number could make that region of the iris map to a 00 some of the time, and a 10 at other times. n this case, we would expect the real bit to be fragile because half ofthe time that bit would be a 0, and half ofthe time that bit would be a 1. Furthermore, we would expect the imaginary bit to be consistent, because the imaginary bit is equal to zero no matter whether the complex number is in the third or fourth quadrants. Figure 9 shows an example of the distribution of 64 complex numbers from 64 different iris images. Each of these 64 complex numbers is associated with the same location in the iris code. n particular, this location on the iris code had a highly consistent imaginary bit and a highly inconsistent real bit. As we expected, the complex numbers associated with these two bits of the iris lay close to the that the outer bands of the iris contained less discriminating texture features than the inner bands of the iris. Since the inner part of the iris is less likely to be occluded, we suspect that it is the occlusion, rather than lack of texture, that might make outer bands less valuable. To test this idea, we graphed the percent of fragile bits that occurred at each band in the iris. Some of the bits in the iris code never changed over all different images ofthe same eye. These bits have probability close to p = 0.0 of changing. Bits that were equal to 0 half of the time and equal to 1 the other half of the time have a probability close to p = 0.5 of changing. We arbitrarily chose to let the threshold for determining which bits are fragile be p=0.4. That is, any bit that flipped less than 4000 of the time was considered "consistent" for this experiment, and any bit that changed more than 400O of the time was considered "fragile" for this experiment. Using this threshold, we calculated the percent ofbits that were fragile for each row of the "unwrapped" iris image, for all 24 iris images. The results are shown in Figure 10. t T+ e imaginary axis 1 T j. j t ~ ~ ~ ~ ~ ~ ~ ~ ~ Row Number (row 1 is next to pupil) 6 Fig 10. This figure shows the percent of fragile bits in each row of the iris code. Rows in the middle ofthe iris code (rows 5 through 12) are the most consistent. Numbers in the legend refer to the subject number in the CE dataset. U0U2 0.0)i rr-nni1ai s LF, -0.01i T1 negative y-axis. U.U4 t t F n fm Fig 9. These 64 complex numbers, each from the same region on 64 different images of the same subject's eye, all correspond to the same location on the iris code. Each complex number is mapped to two bits. This particular part of the iris code had a highly consistent second (imaginary) bit, and a highly inconsistent first (real) bit. Despite our efforts to obtain unoccluded images for the dataset, we suspected that specular highlights close to the eyelids could be affecting our measurements of the consistency ofthe different bands ofthe iris code. Therefore, we applied a mask to the data to mask the upper and lower quarters of the iris, and then we looked at the percent of fragile bits for each row of the iris code in the unmasked quarters of the iris (left and right sides of the iris). The mask used is shown in Figure 11, and the result is shown in Figure 12. Fig 1 1. Mask used for evaluating the consistency of the inner versus V. NNER BANDS VS. OUTER BANDS Several researchers [8,9,10] have suggested using only the inner part of the iris for recognition. Du et al. [ 1] suggested outer bands of the iris. When we consider only the right and left sides of the iris, and disregard the top and bottom sections, which had the most specular highlights and occlusion, there is not as much spread Authorized licensed use limited to: UNVERSTY NOTRE DAME. Downloaded on December 21, 2009 at 15:14 from EEE Xplore. Restrictions apply.

5 in the fraction of fragile bits across the rows. There still seems to be a high percentage of fragile bits in rows 1 and 2 of the iris code; however, the general trend shows that all rows ofthe iris code have a high amount of consistent information. The fragility threshold, p = 0.4 is used for the graphs presented here. We performed the same test using thresholds of p = 0.30 and p = n all cases, masking the top and bottom quarters improved the consistency of the outer rows. Clearly, the quality of the segmentation affects the value of the outer rows of the iris z using bits from all 4 quarters using bits from left & right quarters only 0.25 te U.115 reject rate is FRR(dT) = Pr(h(Q, R) > dt h(q, R) = 0) pi(l1_p)n-i = 1, i =dt +1 where p is the probability that a bit will flip, and (1 - p) is the probability of not flipping a bit. Since we showed that not all bits have the same probability of flipping, we modify Bolle et al.'s equation to include two different probabilities. Suppose that k of the bits have probability p1 of flipping, and N - k have probability P2 of flipping. The first k bits will be termed "Set 1" and the second k bits are "Set 2". Let p, > P2 so that Set 1 contains the more fragile bits. Let i be the number of bits that flip when iris code Q is acquired. The normalized Hamming distance is the fraction of bits in the two iris codes that disagree, so the probability of a false reject is the probability that i/n > dt, or equivalently, that i > N dt. Let j be the number of bits from Set 1 that flip, and let i - j be the number of bits from set 2 that flip. The probability of j bits flipping, out of k bits from Set 1 is G Fowv Number 14 1G J Fig 12. When the regions most affected by specular highlights and occlusion are masked, the consistency of the outer rows of the iris code improves. (The black horizontal line is drawn to aid in comparison between outer and inner rows.) The probability of i -j bits flipping, out ofthe N -k bits in set 2 is V. THEORETCAL MPACT OF FRAGLE BTS ON FALSE REJECT RATE Thus, the total FRR when using all N bits from the iris code is Knowing which bits in a particular subject's iris code are fragile could improve recognition performance. At enrollment, a sequence of iris images could be taken and analyzed to create the enrollment iris code and a mask which would mask out inconsistent bits. Such a mask could be used in addition to a mask for masking out eyelids and eyelashes. This section presents the theoretical false reject rate for an iris recognition scenario, and then shows how this rate can be improved by masking out the fragile bits in the iris code. We try to follow as closely as possible the notation presented by Bolle et al. in their calculations of FRR. We make two different calculations for FRR. The first calculation presents the error rate for a traditional method that uses all bits in the iris code, and the second calculation presents the error rate for a method that masks fragile bits and only uses the more consistent bits. Let Q and R be real world iris codes of length N, both from the same iris. When iris code Q is calculated by the iris acquisition system, some ofthe bits flip so that the result, Q, is only an approximation to the true iris code, and does not match iris code R exactly. f enough ofthe bits flip so that the Hamming distance h( Q, R) exceeds the decision threshold, dt, then a false reject occurs. Bolle et al. showed that the false (P2 ) (1 P2 ) k N-k k(pl)p Y.j(1) j(2) P2)_N-k-i+j summed over all possible values for i and j: i ranges from N dt to N, andj ranges from max{i -N + k, 0 to min{i, k}. Now consider the situation where the k fragile bits from Set 1 are masked out. Let it be the number of remaining bits that flip. The probability of a false reject is the probability that i*/(n - k) > dt. The total FRR when using N - k bits is NE i*=(n-k)-dt N-k pi(_pnki i V. NUMERCAL EVALUATON To compare the situation of using all N bits to the second situation of using only N - k bits, we assume some numerical values. Bolle et al. used N = 173 because Daugman's 1993 paper found 173 independent degrees of freedom in the iris code. n order for our results to be easily comparable to Bolle et al., we also take N= 173. We analyzed the experimental consistency data so we could choose reasonable values for k, Pt, and P2. For each subject, we divided the bits into two groups: fragile bits with probability greater than 0.4 of flipping, and consistent bits Authorized licensed use limited to: UNVERSTY NOTRE DAME. Downloaded on December 21, 2009 at 15:14 from EEE Xplore. Restrictions apply.

6 with probability less than 0.4 of flipping. We counted how many bits fell into each group. On average, 15.0% ofbits had probability greater than 0.4 of flipping, and 85.0% of bits had probability less than 0.4 of flipping. The probability of a bit from the first group flipping averaged , and the probability of a bit from the second group flipping averaged % of 173 is approximately 26, so we let k = 26. We also let pl= and let P2= Assuming these numerical values, the FRR for the traditional scenario is , and the FRR for the scenario which masks the fragile bits is Thus, detecting fragile bits and masking them out can improve accuracy. V. CONCLUSON The consistency of the different bits in an iris code has not been studied by any other previously published work. Our experiments prove the existence of fragile bits. There appears to be no gender difference in the consistency of iris code bits. The evidence is not conclusive as to whether there might be an Asian/Caucasian race difference. There is not a large difference between the consistency of the inner and outer bands, although the inner and outer bands seem to be slightly worse than the middle bands. An important direction for future research is to test how modifications of the algorithm affect the consistency of the iris code bits. This study holds great potential for increasing our understanding of the workings of Daugman-style iris recognition systems, and for significantly decreasing the error rates of such systems. n the future, iris biometrics could potentially be used with extremely large populations. Any application of iris biometrics on a nationwide scale would necessitate extremely low error rates. Based on our numerical evaluations, we expect that we could potentially reduce the false reject rate by three orders of magnitude by using our knowledge of consistent and inconsistent bits. [3] Kevin Bowyer, Karen Hollingsworth, and Patrick Flynn. mage understanding for biometrics: a survey. University ofnotre Dame CSE Technical Report, [4] P. J. Phillips, W. T. Scruggs, A. J. O'Toole, P. J. Flynn, K.W. Bowyer, C. L. Schott, and M. Sharpe. FRVT 2006 and CE 2006 large-scale results. Technical report, National nstitute of Standards and Technology, NSTR 7408, March [5] Ruud M. Bolle, Sharath Pankanti, Jonathan H. Connell, and Nalini Ratha. ris individuality: A partial iris model. n nternational Conference on Pattern Recognition, pages 11: , [6] National nstitute of Standards and Technology. ris Challenge Evaluation, 2006, [7] Xiaomei Liu, Kevin W. Bowyer, and Patrick J. Flynn. Experiments with an improved iris segmentation algorithm. n Fourth EEE Workshop on Automatic dentification Technologies, pages , October [8] Christel-loic Tisse, Lionel Martin, Lionel Torres, and Michel Robert. Person identification technique using human iris recognition. Vision nterface, pages , [9] Li Ma, Tieniu Tan, Yunhong Wang, and Dexin Zhang. Local intensity variation analysis for iris recognition. Pattern Recognition, 37(6): , Feb 2004 [10] Kazuyuki Miyazawa, Koichi to, Takafumi Aoki, Koji Kobayashi, and Hiroshi Nakajima. An efficient iris recognition algorithm using phase-based image matching. nt. Conf on mage Processing, pages 11:49-52,2005. [11] Yingzi Du, Bradford Bonney, Robert ves, Delores Etter, and Robert Schultz. Analysis of partial iris recognition using a 1 -D approach. nternational Conference on Acoustics, Speech, and Signal Processing, pages 11: , ACKNOWLEDGMENT Biometrics research at the University of Notre Dame is supported by the National Science Foundation under grant CNSO , bythe Central ntelligence Agency, bythe US Department of Justice/National nstitute for Justice under grants 2005-DD-CX-K078 and 2006-J-CX-K041, by the National Geospatial ntelligence Agency, and by UNSYS Corp. The opinions, fimdings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of our sponsors. REFERENCES [1] John Daugman. How iris recognition works. EEE Transactions on Circuits and Systemsfor Video Technology, 14(1):21-30, [2] Richard P. Wildes. ris recognition. n Biometric Systems: Technology, Design and Performance Evaluation, pages Spring-Verlag, Authorized licensed use limited to: UNVERSTY NOTRE DAME. Downloaded on December 21, 2009 at 15:14 from EEE Xplore. Restrictions apply.