Iris Recognition Systems and methods Jarkko Vartiainen Lappeenranta University of Technology, Department of Information Technology P.O. Box 20, 53851 Lappeenranta, Finland vartiain@lut.fi Abstract. This is an overview of the new emerging biometric technology called iris recognition. The focus will be on image processing and security aspects. The most known algorithms are introduced and discussed and background for iris recognition are given. Results will show that iris recognition is very good biometric that is comfortable to use for person identification. 1 Introduction Iris recognition is a part of biometric identification methodoligies which also include facial, fingerprint, retinal and many other biological traits. They all offer a new solutions for person identification, authentication and security. Currently users have to carry security badges or know certain pin/pass codes in order to get into secure zones or to log into a computer. Problem with these methods is that users have to remember lots of different passwords and pincodes which therefore tend to be rather easy to guess and crack since users prefer passwords that are easy to remember. Cards can be lost and they can be used by anyone to gain access to a restricted area or to a restricted computer. Biometrics on the other hand provide a certain and easy way of authenticating persons, biometrics are also quite hard to forge and combined with some other method like password they form up a very strong authentication method. Biometric identification utilises many psychological and physical characteristics that define us as a individual. Some more common features are fingerprints, hand shapes, eyes retinas and many others, including eye s iris. Psychical and behavioural characteristics include for example typing speed, walking style and signature. Out of all physiological properties iris patterns are believed to be one of the most accurate. [4] Iris recognition is in many ways a very good research topic in computer science. It combines many aspects of information technology research, such as computer vision, pattern recognition, statistics and human-machine interface. The purpose of iris recognition is real-time, high confidence recognition of a person s identity by mathematical analysis of the random patterns that are visible within the iris of an eye from some distance. Iris recognition has many practical uses, it can be used to authenticate persons identity or to identify a certain person from a large set of data.
2 Background Iris identification methods are quite new in the computational world. The idea that people can be identified by the shape of their irises was first documented in an iophthalmology textbook by James Doggarts in 1949. After that the idea lie dormant for decades until 1987 two ophthalmologists, Aran Safir and Leonard Flom, patented this idea. In 1989 they asked John Daugman to create an actual algorithm for the iris recognition problem. These algorithm that he developed and patented in 1994 form all the basis for the current iris recognition research and products. [7] There has always been interest for the human iris and it has been studied very long time by iridologists. Iridology resembles palm-reading, thus it has no scientific background. Iridologists claim that they can see just by looking at a persons iris the state of his inner organs, health and even his personality. This all was declared as a medical fraud by five different medical journal publications. But as the iridilogists prove, the iris has been in the interest of man long before it became interesting scientifically for computer scientists. 3 Physiology of the iris 3.1 The Iris The iris is a protected internal organ of the eye, located behind the cornea and the aqueous humour, but in front of the lens (see fig. 1). The iris has many features that can be used to distinguish one iris from another. One of the primary visible characteristic is the trabecular meshwork, a tissue which gives the appearance of dividing the iris in a radial fashion that is permanently formed by the eighth month of gestation. During the development of the iris, there is no genetic influence on it, a process known as chaotic morphogenesis that occurs during the seventh month of gestation, which means that even identical twins have uncorrelated minutae, i.e. differing irises. In fact, even persons own eyes are uncorrelated. Pigmentation of the iris on the other hand still continues on the first year after birth. It is usual that new born baby has blue eyes, but after first year the babys eyes may have changed colour. What makes the iris also interesting is the fact that subjects iris is rather easy to photo from a distance in unintrusively and perhaps even inconspicuously. The most important function of the iris is controlling the size of the pupil. Illumination, which enters the pupil and falls on the retina of the eye, is controlled by muscles in the iris. They regulate the size of the pupil and this is what permits the iris to control the amount of light entering the pupil. The change in the size results from involuntary reflexes and is not under conscious control. This feature can be used to guarantee that the image being taken is probably with a high confidence a living eye and not an artificial image of an eye. [10, 4] 2
Fig. 1. The structure of the human eye. 3.2 Features of the iris Among the pigment related features belong the crypts and the pigment spots and naturally the color of the iris. Crypts are noticeable thin lines that extend from the pupil to the edges of the iris. Pigment spots are random concentrations of pigment cells in the visible surface of the iris. They are known as moles and freckles with nearly black colour. Features that control the actual size of the pupil are called radial and concentric furrows. Together they are called contraction furrows and they control the size of the pupil, which in turn controls how much light gets into the eye. [14] Typical radial furrows usually begin near the pupil and extend through the collarette. The radial furrows are creased in the anterior layer of the iris, from which loose tissue may bulge outward and this is what permits the iris to change the size of the pupil. The concentric furrows are generally circular and concentric with the pupil. They typically appear in the ciliary area, near the periphery of the iris and permit to bulge the loose tissue outward in different direction than the radial furrows. Collarette, mentioned briefly above, is the boundary between the ciliary area and the pupillary area. It is a sinuous line as can be seen from figure 3, which forms an elevated ridge running parallel with the margin of the pupil. The collarette is the thickest part of the human iris. [14] The most striking part of the iris if of course the pupil, black round dot in the middle of the iris as can be seen in figures 2 and 3. Pupil may at first glance seem round in shape, but in actuality it may not be exactly circular in shape and its deviation from the circle is a visible characteristics. Centers of the iris and the pupil are different and they can differ from each other of about 20%. All previously mentioned radial and angular variations taken together constitute a distinctive identity that can be imaged from some distance. Further properties of the iris that enhance its suitability for use in high confidence identification systems include its inherent isolation and protection from the external environment. Humans protect their eyes instinctly since they are the most valuable of human senses. Irises are also impossible to be surgically modified without 3
Crypt Collarette (dottet line) Radial furrows Fig. 2. Visible features of the eye. unacceptable risk to vision and iris physiological response to light provides one of several natural tests against artifice. Only rarely the iris texture changes, this is usually due to aging or trauma to the eye, in which case atrophic areas may appear in the iris resulting in a moth-eaten texture. Tumours may grow on the iris, or congenital laments may occur connecting the iris to the lens of the eye. [14, 8, 2] 4 Iris recognition The work of John Daugman is considered as the foundation that all current research is more or less based on. Also all of the commercial applications that are currently in the markets are based on his method [11]. Before Daugman started to tackle with the problem, there were no studies made whether or not irises have sufficient degrees-of-freedom, or forms of variation in the iris among individuals, to use them as a fingerprints to distinguish different persons. It was also unknown whether or not an efficient enough algorithm could be developed to extract a detailed iris images from video image. Also the iris code should be short enough and yet mathematically varying enough that it would be able to render a decision about individual identity with high statistical confidence. And all that should also happen in less than one second of computational time with current general purpose micro processor. [2] 4.1 Daugman s method In Daugman s work the visible texture of a person s iris in a real-time video image is encoded into a compact sequence of multi-scale quadrature 2-D Gabor wavelet coefficients, whose most-significant bits comprise a 256-byte iris code. The iris recognition procedure is a three step process. First image of the eye is 4
Crypts Concentric furrow (dotted) Fig. 3. More features of the iris. captured by using a standard digital video camera. Then from the image, eye and iris are located and finally iris code is calculated and compared to the database. [2, 17] Image acquisition Iris analysis begins with reliable means for detecting whether an iris is visible in the video and then precisely locating its inner and outer boundaries. This is done by utilizing the fact that iris is round in shape and thus by integration and differentation needs to be applied in order to find the correct location. This is accomplished by maximizing the blurred partial derivative, with respect to increasing radius r, of the normalized contour integral of the image along a circular arc of radius r and the iris center coordinates. The complete operator behaves in effect as a circular edge detector. [2] After the iris roughly located a second search finds the fainter pupillary boundary by using a finer convolution scale and smaller search range. The end result is the precise location of the outer boundaries of an iris and the pupillary boundary. Facts like the knowledge that screla is always lighter than the iris are also used to make the algorithm more precise. There are also difficulties like the fact that the pupil isn t always darker than the iris. That problem is resolved by using the absolute value of the partial derivative. This increases the performance of the operator as circular edge detector regardless of these polarities. Also by using near infrared illumination, darker irises will show more details and also user does not see the light which makes it unintrusive.[2] The final codes that will represent the iris have to be extracted from corresponding areas of iris texture. The same regions of the iris need to be tested for similarity. Scaling and the overall iris image can vary due to pupillary contraction 5
or difference in the camera distance. This problem is cicumvented through the use of a projected polar coordinate system and by modelling the irirs as a nonelastic rubber sheet. This model assigns a pair of dimensionless real coordinates (radius, angle) to each point of the iris. [2, 4] After this mapping zones of analysis are defined in this projected doubly dimensionless coordinate system. These zones disregard the top of the iris (because of eyelid coverage) as well as the area where the light source coming from below causes a corneal reflection. The illumination comes from an angle, even if it causes reflections, because it helps avoiding influence of sunglasses. [2] Feature extraction After the pupillary and iris/screla boundaries have been located, any occluding eyelids detected and reflections or eyelashes excluded, the isolated iris is mapped to size-invariant coordinates and demodulated to extract its phase information using 2D Gabor Wavelets. Daugman introduced these filters in 1980. Their mathematical properties include the ability of providing high-resolution information about the orientation and spatial frequency content of the image structure. The demodulation process is shown in figure 4 Fig. 4. The phase demodulation process used to encode iris patterns. The angle of each projection phasor is quantized to its quadrant, setting two bits of phase information. This process is repeated all across the iris with many wavelet sizes, frequencies, and orientations to extract 2048 bits. It amounts to a patch-wise phase quantization of the iris texture, by identifying in which quadrant of the complex plane each resultant phasor lies when a given area of the iris is projected onto complex-valued 2D Gabor wavelets: h {Re,Im)} = sgn {Re,Im)} ρ φ I(ρ, φ)e iω(θ 0 φ) e (r 0 ρ) 2 /α 2 e (θ 0 φ) 2 /β 2 ρdρdφ (1) 6
where h {Re,Im)} can be regarded as a complex-valued bit whose real and imaginary parts are either 0 or 1 depending on the sign of the 2D integral. I(ρ, φ) is the raw iris image in a dimensionless polar coordinate system that is size- and translation-invariant and also corrects for pupil dilation. α and β are the multi-scale 2D wavelet size parameters. They span an 8-fold range from 0.15mm to 1.2mm on the iris. ω is wavelet frequency, spanning 3 octaves in inverse proportion to β. (r 0, θ 0 represent the polar coordinates of each region of iris for which the phasor coordinates h {Re,Im)} are computed. Altogether 2048 phase bits (256 bytes) are calculated for each iris. The number of bytes was chosen according to the capacity of the three channel magnetic stripe of the standard credit cards. This also happens to be the upper bound of the capacity of the iris information. [2, 4] Only phase information is used in the IrisCode(TM) (phase code) since amplitude information is not very discriminating and it depends on exrtaneus information like image contrast, illumination and camera gain. The extraction of phase has also the advantage that phase angles are assigned regardless of how poor the image contrast is as with very poorly focused images. [2, 4] Pattern recognition The key to iris recognition is the failure of a test of statistical independence, which involves so many degrees-of-freedom that this test is virtually guaranteed to be passed whenever the phase codes for two different eyes are compared, but to be uniquely failed when any eye s phase code is compared with another version of itself. This way the problem of pattern recognition is converted to a simple statistical test of independence. In order to reach the recognition result the Hamming Distance of the code of the new iris and all the stored codes is calculated. A simple XOR operation between the corresponding pair of codes provides this Hamming Distance: HD = (codea codeb) maska maskb maska maskb (2) Hamming code measures the dissimilarity between two irises whose phase code bit vectors are denoted {codea, codeb} and whos mask bit vectors are denoted {maska, maskb}. The denominator tallies the total number of phase bits that mattered in iris comparisons after artifacts such as eyelashes and specular reflections were discounted. The resulting HD is a fractional measure of dissimilarity. In the original work Daugman did not use mask bit vectors, later he improved the original algorithm by including mask bit vectors. Masking bits signify whether any iris region is obscured by eyelids, contains any eyelash occlusions, specular reflections, boundary artifacts of hard contact lenses, or poor signal-to-noise ratio and thus should be ignored in the phase code as artifact.[4, 6] Because any given bit in the phase code for an iris is equally likely to be 1 or 0, and different irises are uncorrelated, the expected proportion of agreeing bits between the codes for two different irises is HD = 0.500. Figure 5(a) shows 7
the distribution HDs obtained from 9.1 million iris pair comparisons. The observed mean HD was p = 0.499 with standard deviation σ = 0.0317. Their full distribution corresponds to a fractional binomial having N = p(1 p)/σ 2 = 249 degrees-of-freedom as can be seen from the solid curve. This shows that each comparison between two phase codes bits from two different irises is essentially a Bernoulli trial even though there are correlations among the coin tosses. In any given IrisCode, only small subsets of the code are mutually independent due to the internal correlations within iris. Bernoulli trials still remain binomially distributed but with a reduction in N and thus increasing the deviation of the normalized HD distribution. The theoretical binomial distribution plotted in the figure 5(a) has the fractional form: [4] f(x) = N! m!(n m)! pm (1 p) (N m) (3) where N = 249, p = 0.5, and x = m/n is the outcome fraction of N Bernoulli trials. (a) (b) Fig. 5. a) Distribution of Hamming Distances from all 9.1 million possible comparisons between different pairs of irises in the database. The histogram forms a perfect binomial distribution with p = 0 : 5 and N = 249 degrees-of-freedom, as shown by the solid curve (Eqt 4). The data implies that it is extremely improbable for two different irises to disagree in less than about a third of their phase information. b) Decision environment for relatively unfavourable conditions. Darker histogram shows the distribution of shortest HDs from all 9.1 million possible comparisons after 7 possible rotations. What is also remarkable in the work is the fact that irises that are genetically identical have the same distribution of Hamming distances than non correlated 8
irises. Meaning that persons left and right eye have totally different irises, which also means that identical twins have also totally different irises. [4] Next question that hasn t yet been answered is that how the algorithm copes with the fact that head can be tilted from one side or the other while the iris is being imaged. The solution is to rotate the IrisCode within certain limitations and then make the identification comparison again. After the comparisons only the code that produces the shortest HD is chosen as the matching IrisCode. Even after allowing for 7 different degrees of eye or head tilt the distributions of hamming distances the distribution is only slightly biased toward a lower mean Hamming distance 5(b). The lighter bargraph in figure 5(b) shows the same distribution between pairs of different iris codes given for each given iris allowing again 7 different degrees of eye or head tilt. As can be seen from the figure, the difference between iris comparisons even after some degrees of freedom are still quite clearly visible. Comparison of matching irises and different irises have distinctive difference. Of the 9.1 million different iris comparisons plotted as dark histogram in figure 5(b) the smallest HD that was measured by Daugman was 0.334. The number means that only 2/3 (66%)of the IrisCode bits matched when comparing irises from different eyes. The binomial cumulative from 0 to 0.300 is 1 in 10 billion which is roughly the number of human eyes on the planet. Thus even the observation of a relatively poor degree of match between IrisCodes for two different iris images (say 70% agreement or HD = 0.300) would still provide compelling evidence of identity because the test of statistical independence is still failed so convincingly. [5] Table 1. False match probability as a function of decision criterion. HD criterion Odds of false match 0.26 1 in 10 13 0.27 1 in 10 12 0.28 1 in 10 11 0.29 1 in 13 billion 0.30 1 in 1.5 billion 0.31 1 in 185 million 0.32 1 in 26 million 0.33 1 in 4 million 0.34 1 in 690 000 0.35 1 in 133 000 0.36 1 in 28 000 0.37 1 in 6750 0.38 1 in 1780 0.39 1 in 520 0.40 1 in 170 The statistical data and theory presented above show that iris recognition can be performed successfully by just a test of statistical independence. Any 9
two different irises will most probably pass the test of statistical independence whilst any two images that fail this test (i.e. produce HD 0.32 ) must be from the same iris. Thus it is the unique failure of the test of independence that is the basis for iris recognition. Table 1 shows the probabilities of false acceptance rates based on different HD criterion. [6] There is another quantative way to calibrate the power of decision making for this type of two-choice task by using a metric called d. Decidebility index measures how well separated the two distributions are, since recognition errors are caused by their overlap. If the means of the distributions are µ 1 and µ 2 and their standard deviations σ 1 and σ 2, then d is defined as d = µ 1 µ 2 (σ 2 1 + σ 2 2 )/2 (4) This measure of decidibility is independent of how liberal or conservative is the acceptance threshold used. By measuring separation it reflects the degree to which any improvement in the false match error rate must be paid for by worsening of the failure to match error rate. The measured decidibility for iris recognition is d = 7.3 for the non-ideal condition presented in figure 5(b). The value is higher than any other biometric has ever achieved. [6, 3] These extremely good statistical properties make iris recognition such a good identification method. Verification (one-to-one comparison) as a process is much more simpler than identification (one-to-many comparison). For example if verification method that has 99.9% success rate, is used for identification in a database of size 2000, then the verification method makes a false acceptance in 86% of the cases. Iris recognition on the other hand can adapt the HD threshold and can thus always keep the false accept rate nearly a constant regardless of database size. [5] It is no wonder that all the current commercial iris recognition systems are based on Daugmans work. The algorithm is robust and extremely effective and it has more uses than mere verification, it can be also used for identification which means that no ID badges or keycards are required for gaining access to sensitive areas or information. The iris recognition system patented by Daugman has not failed a single test of acceptance to date even though it has been tested quite exhaustively even with a database of size 984 million template pairs (Test was carried out by Iridian Technologies in 2003, the report is available in their website after registration). 4.2 Other methods There are also other researchers in the iris recognition field even though Daugman is the best known. Wildes et al. [16] introduce a system which differs a bit from Daugmans. First, their system finds the eye of the subject without giving the subject any special feedback like in Daugmans system, where the user sees his eye constantly in a monitor and must move his head in order to get a clear image. Also Wildes light source is diffused and polarized whilst Daugman uses 10
near infrared point light source. Polarization allows Wildes system to ameliorate the effects of specular reflections in the iris imaging. Feature extraction in Wildes system is based on an isotropic bandpass decomposition derived from the application of Laplacian of Gaussian filters to the image data. Matching the obtained and the stored iris representations is based on normalized correlation (NC) between both representations. Let p 1 [i, j] and p 2 [i, j] be the two images arrays of size n m and let µ 1, µ 2 and σ 1, σ 2 be their means and standard deviations. Then the normalized correlation between p 1 and p 2 can be defined as NC = σn i=1 σm j=1 (p 1[i, j] µ 1 )(p 2 [i, j] µ 2 ) nmσ 1 σ 2 (5) the Laplacian pyramid representations instantiate four spatial frequency bands, so four scores are obtained, each accounting for the goodness of match at each frequency band. Finally, it is necessary to combine these four opinions into a single final decision. In an opinion fusion scenario, Wildes chooses to use a Fisher s linear discriminant, applying a threshold afterwards. The weakness of Wildes method is that it s computationally very expensive since it relies on image registration and image matching and most of all it can only be used for verification and not identification. [16] Another researcher [1] has also worked with iris recognition. His approach is based on calculating the zero crossings of the wavelet transform. 2002, In his work apart from the iris location a normalization algorithm brings the iris to have the same diameter and the same number of data points. From the grey levels of the sample images, one-dimensional signals are obtained and referred to as the Iris signature. Then a Zero Crossings representation is calculated based on the Wavelet Transform. These representations are stored as templates and are used for the matching algorithm. In this way the author claims that the noise influence will be eliminated since zero crossings are not affected by noise. Interesting aspect in his work is the ability of the wavelet transform to eliminate the effect of glares due to reflection of the light source on the surface of the iris. This was a problem that neither Daugman nor Wildes were able to solve. Also he tested various resolutions and chose the most significant levels which contained most of the energy of the iris signature and thus were less affected by noise. [1] After locating the pupil with the assumption that it composed a circular closed contour the centroid of the pupil is chosen as a point of reference. Using this reference, concentric circles are formed to collect data in circular buffers. From each buffer an Iris signature is generated. This procedure needs an additional normalization process since the diameter of the iris could vary. In order to achieve this, the maximum diameter is chosen as reference and is used to scale the virtual circle diameters to constant size. Normalization also takes place with the data points. Since the same number of points are needed, a normalization value is selected to help the wavelet transform to extract all the information available in the iris signature. [1] 11
In order to reach the recognition result, models of iris signatures are made from all the signatures using the same normalization constants for both the users irises as well as the testing irises. Using the number and the location points of the zero-crossings, a dissimilarity measure is obtained. The iris with the minimum value is chosen as the correct target. The dissimilarity value is the average of the respective dissimilarity values at the various resolution levels. The first two are calculated using all the data points and the final two using only the zerocrossings. However, further processes are needed to compensate for different number of zero-crossings. [1] There are also other methods [11, 13, 12] that have been invented, but in general it can be said that they all are inferior in comparison against Daugmans method. There is still room for improvement in the field and all the methods given here need improvement before the system is ready to be used very widely. 5 Reliability of iris recognition 5.1 Possible ways to trick iris recognition systems There are several ways to confirm that a living iris is being scanned and not for example a photograph, a videotape, or a fake iris printed on a contact lens, glass or other artifice. One way is of course to measure the ratio of pupil diameter against to iris diameter either when light levels are changing or even under steady illumination. The pupil size can be controlled by preprogrammed random changes in the light level with a response time constant of about 250ms for constriction and about 400ms for dilation. But even without programmed illumination changes, the disequilibrium between excitatory and inhibitory signals from the brain stem to the enervation of the pupillary sphincter muscle produces a steady-state small oscillation at about 0.5Hz termed hippus. Since the algorithms usually has to track both the pupillary boundary and the iris boundary, it is routine to monitor the amount of hippus. Its coefficient of variation is normally at least 3%. [5] Other ways to exclude a photograph of somebody else s iris involve tracking eyelid movements, or examining corneal reflections of infrared LEDs illuminated in random sequences. Still further measures could test for the characteristic spectral signature of living tissue in infrared illumination. Hemoglobin in oxygenated blood has an absorption band in the near infrared wavelengths, whereas printer s dyes and emulsions and reflectance properties of photographic papers are often completely ineffective for infrared light. [5] Finally a person can try to fool the iris recognition system with a contact lenses with faked iris pattern printed on them. The fact that such a fake iris is floating on the spherical, external surface of the cornea, rather than lying in an internal plane within the eye, lends itself to optical detection. Also, the printed iris pattern doesn t undergo any distortions when the pupil changes in size, as does a living iris pattern. Moreover, the printing process itself creates a characteristic signature that can be detected, as can be seen in figure 6. The 12
image shows a natural iris and a fake one printed onto a contact lens and their 2D Fourier power spectra. The dot matrix printing process generates four points of spurious energy in the Fourier plane, corresponding to the directions and periodicities of coherence in the printing dot matrix, whereas a natural iris does not have these spurious coherences. [5] There are some problems with the iris recognition too, the object to be tracked is very small within moving target and is obscured by eyelashes, lenses and reflections. Iris is located behind a curved, wet reflecting surface and it also deforms non-elastically, which make the pattern tracking rather challenging task. Illumination should also not be too visible or bright or it annoys the users and eventually when technology goes far enough it offers some negative Orwellian connotations since every people could be rather easily tracked by their eyes, since we have to use them for almost everything. [9] 5.2 Testing of some commercial iris recognition system Since biometrics are becoming more and more general, a German research group tried to defeat several biometric systems by means that are widely available. Iris recognition system was thought to be the hardest to defeat. First they tried to offer high resolution images of an iris via notebook notebook pc display as well as via a head-mounted display with no success. Also regular paper offered no results, but by looking at the images the system took, they noticed a bright spot in the middle of the pupil which gave them an idea. They cut a small hole in the middle of the paper and the printed the iris image on that same mat paper with resolution of 2400 1200dpi. Then they tried to log into a system with the image by covering a living eye with the image so that a living pupil was in the middle of the image (figure 7). This was enough to convince the system that a living and authentic person was trying to access the system and granted access under the assumed identity of Master False Eye. [15] They also tried to enrol into the system with the aid of an artificial eye. From that point onward, anyone with the possession of the eye pattern was able to log on to the system. Moreover, the person whose eye had been used to create the pattern was also able to acquire authentication in relation to the picture-generated reference data set with his own live iris. [15] It has to be said if favor of the iris scanner that under real life conditions it would be hard to obtain iris images of authorized persons with high enough accuracy. But should such an image be available, then creating a deceptive eye patch would not be much of a problem since high resolution ink jet printers and good quality papers are widely available. So even if in theory a living is easy to identify, the commercial systems still need some work before they can be used for real security purposes. [15] 6 Summary When we need to know with certainty who an individual is or whether he is who he claims to be, we normally rely upon what he possesses like a key or a card or 13
Fig. 6. Illustration of one countermeasure against subterfuge: detecting a printed eye on a contact lens by the 2D Fourier plane artifacts of printing. 14
Fig. 7. Achieving authentication with someone else s iris by hiding your own pupil behind it. we rely on the fact that he knows something that nobody else knows, like a password or a pincode. Other way to identify a person is to use a unique biological trait, something which is unique from person to person, like a fingerprint or appearance. The first two methods are easy to implement technologically and easy to confirm automatically, but they are also very unreliable since everyone can use password or key without proper authorization. Today we know, that person uniqueness is determined by his genes, but DNA testing is not unintrusive nor fast and thus is not acceptable in everyday usage. The remaining options are characteristics that are unique for every person regardless of aging and other aspects. Iris recognition offers very reliable way to recognize and identify persons. Iris is well protected, immutable, internal organ of the eye, that is readily visible externally and has very random patterns from person to person. This identification power, which means that users need not even bother to assert an identity, is one of the main advantages of iris recognition as a biometric. References 1. W.W. Boles. A security system based on human iris identification using wavelet transform. Engineering Applications of Artificial Intelligence, 11(1):77 85, Feb. 1998. 2. J. Daugman. High confidence visual recognition of persons by a test of statistical independence. IEEE transactions on Pattern Analysis and Machine Intelligence, 15(11):1148 1161, Nov. 1993. 3. J. Daugman. Iris recognition. American Scientist, 89(4):326 333, July-Aug. 2001. 4. J. Daugman. How iris recognition works. In 2002 International Confrence on Image Processing, volume 1, pages 33 36, Sept. 2002. 5. J. Daugman. Demodulation by complex-valued wavelets for stochastic pattern recognition. International Journal of Wavelets, Multi-resolution and Information Processing, 1(1):1 17, 2003. 15
6. J. Daugman. The importance of being random: Statistical principles of iris recognition. Pattern Recognition, 36(2):279 291, 2003. 7. J. Daugman. History and development of iris recognition. [html-doc.], [retrieved 10.10.2003]. From: http://www.cl.cam.ac.uk/users/jgd1000/history.html. 8. J. Daugman. Anatomy and physiology of the iris. [htlm-doc.], [retrieved 15.10.2003]. From: http://www.cl.cam.ac.uk/users/jgd1000/anatomy.html. 9. J. Daugman. Advantages of the iris for identification. [html-doc.], [retrieved: 24.10.2003. From: http://www.cl.cam.ac.uk/users/jgd1000/addisadvans.html. 10. J. Daugman. Anatomy and physiology of the iris. [html-doc.], [retrieved: 24.10.2003]. From: http://www.cl.cam.ac.uk/users/jgd1000/anatomy.html. 11. Y-P. Huang, S-W. Luo, and E-Y. Chen. An efficient iris recognition system. In Proceedings of the First International Conference on Machine Learning and Cybernetics, volume 1, pages 450 454, Nov. 2002. 12. Ma L., Y. Wang, and Tan T. Iris recognition using circular symmetric filters. In Proceedings. 16th International Conference on Pattern Recognition, volume 2, pages 414 417, Aug. 2002. 13. L.W. Liam, A. Chekima, L.C. Fan, and J.A. Dargham. ris recognition using selforganizing neural network. In SCOReD 2002. Student Conference on Research and Development, pages 169 172, July 2002. 14. A. Muroň and J. Pospíšil. The human iris structure and its usages. [html-doc.], [retrieved 15.10.2003]. From: http://publib.upol.cz/ obd/fulltext/physica39/physica 39 07.pdf. 15. L. Thalheim, J. Krissler, and R.W. Ziegler, P-M. Translated by Smith. Body check: Biometrics defeated. [www-article], May [retrieved: 24.10.2003]. From: http://www.heise.de/ct/english/02/11/114/. 16. R.P. et al. Wildes. A system for automated iris recognition. In Proceedings of the Second IEEE Workshop on Applications ofcomputer Vision, pages 121 128, Dec. 1994. 17. G. Williams. Iris recognition technology. IEEE Aerospace and Electronic Systems Magazine, 12(4):23 29, April 1997. 16