Biometric Systems BIOM 426
Popular in Course
verified elite notetaker
Popular in Information technology
This 114 page Class Notes was uploaded by Karli Wunsch on Saturday September 12, 2015. The Class Notes belongs to BIOM 426 at West Virginia University taught by Natalia Schmid in Fall. Since its upload, it has received 54 views. For similar materials see /class/202794/biom-426-west-virginia-university in Information technology at West Virginia University.
Reviews for Biometric Systems
Report this Material
What is Karma?
Karma is the currency of StudySoup.
Date Created: 09/12/15
Infrared Identi cation BIOM 426 Biometrics Systems Instructor Natalia Schmid 331mm 33 323133 3quot 3333 133 November 1 2005 WVU Outline Introduction Comparison IRID to other biometrics Principles of IRID Processing Approaches IRID Systems References November 1 2005 WVU Introduction November 1 2005 Thermography is an imaging method using cameras sensitive in the infrared specuunr Surveillance applications Visible and infrared images gait passive imaging distance Thermograms drawbacks effected by ambient temperature by ingestion of certain chemicals medications physiological conditions as inflammation arterial blockages etci Radiometric IR camera can be used to produce EKG WVU 3 Introduction What is used for IRID Anatomical information structure of blood vessels Infrared Cameras passive collecting energy Long 812 micron band and Mid 35 micron band image vessels 4 cm below the surface Two major drawbacks 1 Camera cost 2 No large databases November 1 2005 WVU 4 Comparison to Other Biometrics Uniqueness There is no proof that any biometric is unique Facial Thermography is a robust biometric Thermal patterns are derived from pattern of blood vessels transporting warm blood Person s anatomy does not change aside from growth injury surgical intrusion Identical twins have different thermo grams Thermograms contain more information than fingerprints minutiae 16 minutia points has to match for ngerprints Forgery can detect attempted disguise Image can be blocked but not changed November 1 2005 WVU 5 Figure 92 Visual and infrared images 0 lhree individuals November 1 2005 WVU Comparison to Other Biometrics Example Temperature distribution over artificial hair is different from normal hair distribution Skin surface can be distorted This does not add minutiae Plastic surgery can distort skin but does not change the major vessels Distortions are modeled using rubber sheet effect Thermogram can be distorted but it contains an evidence of this A single frame is used for IRID Different emissivity materials may change thermogram November 1 2005 WVU 7 Figure 95 Visiblr and inl mrcd imagery nl39nn individual having dark umnplcxiun November 1 2005 WVU Comparison to Other Biometrics Uncontrolled Environment IRID works in complete darkness contrary to visible imaging Changes in Appearance high permanence weight gain or loss causes rubber sheeting distortions Nonintrusiveness All individuals provide useful images while may not provide good fingerprints etc Cooperation is not necessary Identi cation of DarkSkinned Individuals Visible light datasets do not contain data of darkskinned individuals November 1 2005 WVU 9 Iemnoxul Anmv Colonial van supmucu Ianwnlvcln Angulm vum Foclul Vuln Inlemnl Juuulal Voln e Mcallnlv Mow ml Jammy Veln Frch Amy J39 Inward Vain v superb 0 Cum Cummun Cmch Allavv madam m MUulv Amy allu meanmv Mom Emch Avlcly Alemy 7 cunnuuc Venn u r Dcscond ng Anna manov Vane Cava Figure 96 Facial and thoracic arteries and veins November 1 2005 WVU Comparison to Other Biometrics Scaling Issue Underlying features locations of speci c junctions among blood vessels 01 inch in diameter Minutiae can be extracted from thermal contours low sensitivity cameras or from absolute location in imagery high sensitivity cameras 175 minutiae can be extracted from full facial image The number of possible configurations exceeds human population Throughput Can provide the fastest throughput noncontact done at a distance does not depend on lighting effects can recognize multiple faces simultaneously Matching is a limiting factor 1 sec for verification November 1 2005 WVU ll Comparison to Other Biometrics False Negatives IRID systems are expected to falsely reject only individuals who have new gross thermal condition sunburn or facial surgery False Accepts Depend on quality of imaging Prior matching cold pixels background hair eyeglasses are converted to random noise Correlation matching of image or its parts Correlation of Visual and Infrared Images matching methods can use visible data bases common features positions of nose eyes hair head shape IR lacks color of the skin hair and eyes provides detailed anatomical information For each IR image system can eliminate all visual images that do not match 95 November 1 2005 WVU 12 Processing Method 1 PCA technique R Cutler 96 Camera 12 bit 25 55 nm infrared range resolution is 160Xl20 Data 24 individuals frontal 45 degrees profile views two expressions normal smile two images 4 sec apart per expression per view no glasses Performance 97 correct classification training set 24 normalexpression images testing set 24 smilingexpression images frontal and 45 degree views one individual is misclassified profile view no misclassification November 1 2005 WVU l3 Processing November 1 2005 Images of 24 subjects WVU Results November 1 2005 Eigenfaces WVU Processing Method 2 Minutiae based F Prokoski 02 Applications Video encoding lRlD Classification Medical imaging General features hot patches in the sinus areas relatively cool cheeks cold hair Individual characteristics specific thermal shapes in certain areas F 5 Cameras FlG la sens1t1ve to wavelength in the 35 Visible and infrared images 812 or 215 micron ranges contain minutiae thermal resolution of 0025 degrees spatial resolution of 002 in 30 maps per second November 1 2005 wvn 16 Processing no an m L mam Visible images with marked minutia points November 1 2005 W V 17 Processing Thermal images November 1 2005 Types of minutiae 1i Brunch points of superficial blood vessels 2 Centroid of each constant thermal area 3 Lymph nodes glands 4 Head outline and hairlines 5i Scars tattoos and other marks 6 Zero crossing of wavelet transform 7 End points where blood vessels goes too deep Approach Use thresholding and consider all hotter area to determine minutiae Previous work Tal et al US Patent no 4975969 Suggests to use landmarks such as head outline hairlines the center of each nostril pupil spacing and the comer of each eye for identi cation Variations in these measurements for a given individual is larger than variations between persons Infrared images in general contain more details overcomes the problem of makeup partial face can be used for identi cation November 1 2005 WVU l9 Processing Steps Comparison and Alignment 1 Face axes are located approximate alignment frontal IR face images are symmetric the canthi and sinus areas are the hottest horizontal axis is drawn through the pupils or canthi other feature is nostrils vertical axis is drawn through the center of the upper lip to the midpoint between the eyes 2 Images are scaled to a standard size standard distances between speci c minutiae The shortest line between canthi 3 cm 3 Infrared minutiae subdivision absolute directly extracted derived image transformation November 1 2005 WVU 20 Methods to derive minutiae Infrared 1 Absolute minutiae as head outline hairlines branch points and end points of superficial blood vessels 2 Derived minutiae a The centroid of each constant thermal area When image has N bits of grey scale begin with dividing image into two slices by thresholding The thresholded image has areas of constant intensity Evaluate a centroid for each of these areas Centroid minutia point xyz xy location wrt center of face Proceed with slicing and deriving new minutiae b Points of maximum curvature on constant thermal contours c Zero crossing points of wavelet transform 1 All pixels above a selected threshold and all pixels within some range Visible All absolute Head outline hairlines pupils eye in and outer comers nostrils mouth comers lip bow and tip of nose November 1 2005 WVU 21 Further Processmg 4 Tables of minutia points are created and coincident minutiae are counted 5 Distance between coincident minutia points is evaluated graph matching bounding box method any other method for matching fingerprint minutiae 1 6 If distance exceeds some threshold the me 5 Fit Em spectrum dependent minutiae are compared Exclusion Test no vascular structure can fall outside the head outline or inside the eye mouth or nostril area Anatomical Rules facial vein and artery must lie outside nose boundaries must not go through mouth eye or nostril areasr Matching minutiae November 1 2005 WVU BlockDiagram Image Acquisition Standardimlion Evaluawr Novembex 1 2005 BlockDiagram Captures image of unknown individual Processing face axes are found IR minutiae are located quality measure is assigned Classification techniques PCA minutiae metrics etc Determine rotation tip and tilt Database is generated infrared visible hyperspe tral medical Images are scaled GENERATE mm UP IMaiGEE 5 IR minutiae are located on the PROCESS FCl 39 a CTR images of other EEEEWEHF 9 FE amass type axes are ass1gned PRCCE 55 I m minutiae are REFERENCE A counted Images are aligned Error bands for minutia are established November 1 2005 WVU Selected images of good quality are stored thresholding 24 BlockDiagram November 1 2005 a a Em WWW 30 name THREE 3 mam VIGLAmN TEST Fm AHATWICM 3quot EULV 36 43 mwm PGTEHTIAI JsflATCHES WVU 25 References 1 F J Prokoski United States Patent No 6496594 B issued on December 17 2002 Method and Apparatus for Aligning and Comparing Images of the FaCe and Body from Di erent Imagers Washington DC US Government Printing Office 2002 2 D A Sokolinsky L B Wolff J D Neuheisel and Ch K Eveland Illumination Invariant Face Recognition Using Thermal Infrared Imagery Computer Vision and Pattern Recognition 2001 CVPR 2001 Proceedings of the 200 IEEE Computer Society Conference on vol 1 pp 527 534 2001 3 R Cuttler Face Recognition Using Infrared Images and Eigenfaces Technical Report November 1 2005 WVU 26 Iris Recognition BIOM 426 Biometrics Systems Instructor Natalia Schmid October 18 2007 WVU 1 Outline Anatomy Iris Recognition System Image Processing John Daugman iris localization encoding Measure of Performance Results Other Algorithms Pros and Cons Ongoing Work at WVU References October 18 2007 WVU 2 Anatomy of the Human Eye Eye 2 Camera Comea bends refracts and focuses light Retina 2 Film for image projection converts image into electrical signals Optical nerve transmits signals to the brain October 18 2007 WVU 3 Structure of Iris cowau MW f V Iris 2 Aperture n WWqu Different types of muscles the sphincter muscle constriction radial muscles dilation Iris is at Color pigment cells called melanin The color texture and patterns are unique October 18 2007 WVU 4 Individuality 0f Iris Left and right eye irises have distinctive pattern October 18 2007 WVU 5 The Most famous Story A LIFE REVEALED The remarkable story of Sharbat Gula first photographed in 1984 aged 12 in a refugee camp in Pakistan by National Geographic photographer Steve McCurry and traced 18 years later to a remote part of Afghanistan where she was again photographed by McCurry is told by National Geographic in their magazine April 2002 issue and on their website John Daugman proved that the above portraits show the same person by running his Iris Recognition algorithms on magnified images of the eye regions in the 1984 and 2002 photographs October 18 2007 WVU 6 The Most famous Story October 18 2007 W V 7 The Most famous Story October 18 2007 W V 8 The Most famous Story October 18 2007 WVU 9 Iris Recognition System in Movies I Rx MH INCREDIBLE v quot J October 187 2007 WVU 10 Iris Recognition System in Movies MINORITY REPORT NS l JUNE 21 l ll ll ll l Will it lll l39tll l l l 7 7 mm H mm 7 m m 7 r rquot o rt gown at Ma w m tr October 18 2007 WVU ll Iris Recognition System Acquisition Image Localization lrisCode Gabor Filters Polar Representation Demarcated Zones October 18 2007 WVU 12 Iris Imaging Distance up to 1 meter Near infrared camera Mirror October 18 2007 WVU Imaging Systems wa39kuquot came39a httpwwwiridiantechcom Salzcled Hardware Partners OK R SFASSVWG Funy Autnmatm Tm Eye 7 Iris Recogm tian Gamer l a a H1724 away operates 2560 m Panasonlc LG Electronics sl llli mns Ftu Autuma t Tm Eye 0K1 r Iris Recogm n en Camera pam m E M39E uu operates at mew away 25760 cm Panasnmt EMEEUD High Speed Fixed Fucus Tm Eye n s Recogni un Camera operates at 1216quot pending away EU40 Em One Eye In LG 3mg Recugmtmnfiantera Operate a 410 away Vaezs Cm October 18 2007 WVU Imaging Systems De ktOP cameras httpwwwiridiantechcom Tethered PC mg Selected Software Partners Recognitionr Videoconferencing F CameraJ operates at 1 b SO ware Inc mmVMMIM 1921quot awa r 4753 cm COMWKBYASMMWS F39 Netegrity n k quotmemanlninneBulloau noun OKI IRISPASSh Tethered F39C Iris V Recognition Camera q operates at 1quot3quot Pending away 339 cm Securimetr icsr lVL PIER 22 Standalone or tethered iris recognition camera Pending October 18 2007 WVU Imaging Systems SARNOEP Corporation Iris on the MoveTM Current commercial systems require relatively close proximity to the eye and significant cooperation from subjects they also impose constraints on the subject s position and movement causing inconvenience and bottlenecks in many situations IOM brings iris recognition up to speed IOM can identify up to 20 subjects per minute and subjects can walk through a recognition portal at normal walking pace IOM s application possibilities include access control Visitor management time and attendance monitoring and border control p 391 L 77 Sarnoff Corporation announced the release of Iris on the MoveTM at the Biometric Consortium Conference on September 22 2005 October 18 2007 WVU l6 Quality Evaluation for Iris 39 r 35 4am L 3 Images from an OKI camera collected at WVU Twostep procedure Evaluation of individual Factors defocus motion blur pixel count occlusion specularities lighting offangle Combining factors using Dempster Shafer fusion technique A09 gm SEW 05 000 o o o O gt 040 0 E03 0 0 o 00 O 3 O 02 63000 0 Z i 03910 100 200 300 400 500 600 700 800 00 560 who who 20 00 2500 Image Number Image Number CASIA Quality per Image Quality per Image WVU October 18 2007 WVU 17 0 Image Processmg John Daugman 1994 Pupil detection circular edge detector maXG0rgtIlti J Ma s V xoayo a 27W V xoayo Segmenting sclera r 7Z398 maX IIp6pdpd6 1 1 rd 5m Omar pZHS 75 r a Hr8 October 18 2007 WVU Rubbersheet Model Each p1xelxyls mapped mto polar pan x o Cuwlarband ls dwxded mto x subbands ofequal thickness for a gwen angle 0 Subbands are sampled uniformly m o and m x Samplmg avemgmg over a patch of pixels Encoding 27D Gabor lterm polar coordinates 000 expi zzmxar 00 r raawrm g Gr9e rm 0760 7 H y 17 cosmmm 7 90 isin727rm9 7 9013 a2 Euler s complex number formula 8 cosx i sinx Oembulx m wvu Encoding ltr ro2 36 60 Gaussian Component 2 b2 6 d a1 7209 October 18 2007 W V Encoding Sinusoidal Component COS 27TwH Ho 0r a1 7209 r020 490 01 SINE October 18 2007 W V sin 27239u6 60 Encoding 704025979le 2 Real Pan cos 27ra6 60e b2 F a 1 E b 09 3 Eu r0 i 90 g a 1 E O 0 lt 0 October 18 2007 WVU 23 Encoding rim wisw Imaginary Pan isin 27mt9 906 2 b2 a1 g A E b 09 D 0 r0 9 DC 60 0 E 4 E a 1 1 0 0 lt 0 October 18 2007 WVU 24 IrisCode Formation Setting the Bits in an IrisCode him 1 if RelAmt nivleilmia r39l eit n lzl 39up wdpdqg 2 0 he u if Help Aewl n lemAFVn e an M M39mepdpd y lt 1 km 1 ifMLAgle v emP1quot edamW qupywpdpdgg 2 0 Am 0 ifImL1peirwtf ni ledmvl lu eianewVHFWMM lt a Intensity is left out of consideration Only sign phase is of importance 256 bytes 2048 bits October 18 2007 WVU 25 Measure of Performance Hamming distance standard measure for comparison of binary strings 1 n D Z Z xk quot9 y k 1 kl x and y are two IrisCodes ED is the notation for exclusive OR XOR Counts bits that disagree XOR 163120 Example 100001100011111 163021 101010000011101 063121 063020 001011100000010 October 18 2007 WVU 26 l3 Measure of Performance contd If there is occlusion exist 1 D Z x G amp mask amp mask sumsummaskx amp masky Z Y x y X and y are two IrisCodes maskX and masky are two IrisCodes 9 is the notation for exclusive OR XOR masks 111111100000111 amp111111000111000 111111000000000 Iriscode 100001100011111 101010000011101 001011100000010 001011000000000 October 18 2007 WVU 27 Observations 39 Two IrisCodes from the same eye form genuine pair gt genuine Hamming distance 39 Two IrisCodes from two different eyes form imposter pair gt imposter Hamming distance 39 Bits in IrisCodes are correlated both for genuine pair and for imposter pair 39 The correlation between IrisCodes from the same eye is stronger Strong radial dependencies Some angular dependencies October 18 2007 WVU 28 14 Observations Read J Daugman s statement with caution Interpret correctly Independence of Bits across IrisCodes The fact that this distribution is uniform indicates that different irises do not systematically share any common structure Probabilily no 0102 03 04 05 06 07 08 0910 a WV 4 39VUWW v AtAmVJl axl iwklel mu MAM n1 WWW WWW For example if most irises had a furrow or crypt in the 12 o39clock position then the plot shown here would not be at 1 u 60 0 Code Bil Localion x 100 120 URL httpwwwclcamacukusersjgd1000independencehtml October 18 2007 WVU 29 Training Sets M users N22 iris images per user Genuine Set usermiris1usermiris2 m 1M Compute M genuine Hamming distances October 18 2007 w M 10000 Imposter Set Formed from combination of irises from different users userki1is1userliris1 userki1is1userliris2 userki1is2userliris 1 userki1is2userliris2 kis not equal to l kl1M Compute imposter Hamming distances WVU 30 Count 300000 500000 100000 Degrees of Freedom lrnposter matching score Binomial Distribution of 91 million IrisCode HDs normalized histogram Solld curve binomial PDF N249 degreesrofrfreedom p05 approlenation curve 9060003 drfferent Ins comparisons Binomial with 249 degrees of freedom Interpretation Given a large number of irnposter pairs The average number of distinctive bits is equal to 249 00 01 02 0 3 04 05 06 0 7 08 09 10 Hamming Distance October 18 2007 WVU Histograms of Matching Scores Decidability Index d prime Decision Environment for Iris Recognition d prirne 1136 same different The cross over point is 0342 mea 01 4 458 strd dev 1 165 stnd dev 1 0197 a Compute FMR and FRR for c 8 every threshold value 25 million comparisons d39 75 0 0 0 1 0 2 0 3 7 0 8 0 9 1 0 Hamming Distance October 18 2007 WVU 32 16 Decision The same eye disliibulions depend slIongly on lhe quality of imaging Nonideal conditions Declslnn Envlmnmem lur Ins Hemgmnon Nonldeal Imagmg r molion him 7 focus 7 noise 7 pose vatiation r illuminalion Demlly m u z mlhan mrmausum ngmm October 1811307 WVU 33 Dec1 0n Ideal conditions Deckson Envymnmenl rm lns Rncognmmr Ideal lmagmg Imaging quality detelmines how L L 39 39 disliibulion evolves and migtates le watdsi m amm deptime fox ideal imaging Dervle deptime 14i1 deptime fox nonrideal imaging ptevious slide um WWW w l M u u as m l u w 39u d PI me7 3 Hammmg Dlslmzce WVU 34 0mm 12 21307 Error Probabilities HD Criterion False False 028 1 in 1012 1 in 11400 1 in 1011 1 in million Crossover 1 in million 1 in million 037 1 in 1 in 113 million Biometrics Personal Identification in Networked Society p 115 October 18 2007 WVU 35 False Accept Rate For large database search Binomial Distribution of 91 million IrisCode HDs FMR is used in verification FAR is used in identification g Soiid curve binomiai PDF N249 degreesrofrfreedom p0 5 N O 9060003 differentiris com arisons FARzl l FMR zNFMR quot 0 a mean 0 499 stnd dev 0 0817 min 0334 max 0664 Adaptive threshold to keep FAR fixed g 00 01 02 03 Hmingzistne 07 08 09 10 cnt OgIO October 18 2007 WVU 36 18 Test Results 7 9mm Lubx SA mum nmm Tum Luisa in uusm Smmu cm l39SA mmm lull Emlmmk L 2mm Eankvr LSA mun Nrmuml mlNrHl Ln n mum l Duugmmt Fix 2mm ludmn Tu mmlrwe 34 mm httpwwwcl camacukusersjgdl 000iristestspdf October 18 2007 r t Aim The results of tests m H published in the period from 1996 to 2003 u Be cautious about reading mm these numbers 19900 m V The Huddle column shows mm H the number of imposter pairs tested not the 27 million n number of individuals per dataset 91 million H 984 mllliou U WVU 37 Performance Comparison amine Face2 39DFPvnmp ercmmzy ownpncal Hand 0 m AIMem vmne Fame Rem Rate False ACCEPL Rate UK National Physical Laboratory test report 2001 httpwwwcl camacukusersjgdl 000NPLsummarygif October 18 2007 WVU 38 Performance Comparison Best0f 3 error rates Fane FF ehgtp nmplt2 FPrnplmsl nitHana Olns AIMem Vame 3 a a 1 False Accept Rate UK National Physical Laboratory test report 2001 October 18 2007 WVU 39 Other Systems R Wildes et 31 System 1 Image Acquisition 6 pixels across diameter 20 cm distance diffuse source circular polarization and a lowlight level camera 2 Iris Localization image is transformed into a binary edgemap contour tting using Hough transforms 3 Pattern Matching alignment of two patterns representation a Laplacian pyramid goodness of match estimate of correlation coef cient Fisher s linear discriminant Iris Recognition An Emerging Biometric Technology Proc of the IEEE 1997 October 18 2007 WVU 40 20 Future of Iris CIA to capture in recognltion at a dlstance r h um Zakaria Wednesday a Navemher 2mm The u s Central Intelllqence ndeney ys develavmq technalaqv that wyu be able a ydenwy Deavle nam their m5 7 eyen wnne they are mavmq at a distance Andrew may 52mm physical scientist at cxn s Intelllqence Technalmjv and Innavatmn Centre tald a c Str ma Interns Hal 2 y y hmmemcs held yn Washinqtan D c this week thatthe n l m e ym r l H e l y can am l y Differs ees Simple femurs like lighting and cdn expressian can ympede identi catmn af sameane uimq yenes an e Pusanm sesnd n currentfamal neeadnman technalmjV sayd Kirby naneays sesnneysnd nne daene eye wavele NAMrm oeenspdn ene yns yeeadnynan eeennaiady Thase differences are sa Slqm cantthat my awn Picture a en yn wa dy enenc Places attwa different times ys actuallv mare dif cuan match than ye wauld he a match me with sameane n this addyenee ne tald the farum way sayd nys Dragram wnyen was created Wu Years ada nas set a nal af ympnayynd aee neeadnman echnaquV by a femur af m Currenth r5 ee dn an H r m e chnalmjv 5 mare reliable than face neeadnman echnalaqv but its limitatmn ys that ye ned ne a ea a alive s y m stand yn frant anne scanner and d ye pnaperiy he ye e b were yaawnd a renaee Kirby sad 39 adqu that it wauld he mare valuable if the in end be awed by a wnne ene was n Man a a dyseanee ea naye the ydeneyneawan One anne mam thrusts af aur pnadnam ys yn factm make M5 Dassihle httpWWWabcnetauscienccncwsstoriess982770htm October 18 2007 WVU 41 Work at WVU 1 Acquisition and Processing of NonIdeal lris Data Li Homak Ge Fahmy X Li Ne Schmid and Si Schuckers Video sequence of face from O to 90 degrees special filter detecting pupil to localize eye 45 lowresolution angular iris images interpolated into 1 high resolution image affine transformation new encoding technique performance evaluation amount of information in nonideal iris 2 Performance Prediction for a LargeScale Iris System B Cukic Ni Schmid and Hi Singh estimation of bit correlation modeling iris data realistic upper bound on performance of irisbased identification system October 18 2007 W V 42 21 Work at WVU 5quot Nonideal Iris Segmentation and Algorithms N Schmid s Schuckers L llomak and x Li 7 relies on quality mctors r compensation of 5 mctors r ellipse fitting 7 ordinal binary coding refinement of undetermined bits 4 Iris Quality using DempsterrShafer criterion N Kalka J Zuo N Schmid B Cukic 5 Generation of Synthetic Irises N Schmid B Cukic and A Ross Zoomed Iris 3 Schuckers N Schmid and L Hornak October lmom WVU 4 Spoof Detection 1 Hipplls steadyrstate small oscillations ofpupil size at about 05 Hz 2 Tracking eyelid movements quotM r 3 Examining ocular re ections 4 optical surth 7 4 re ections 4 2D Fourier spectra printer s dot in artificial irises October 131007 wvv 44 Some References 1 J Daugman s web site URL httpwwwclcamacukusersjgd1000 2 J Daugman High Con dence Visual Recognition of Persons by a Test of Statistical Independence IEEE Trans an Pattern Analysis and Machine Intelligence vol 15 no 11 pp 1148 711611993 3 J Daugman United States Patent N0 5291560 issued on March 1994 Biometric Personal Identi cation System Based on Iris Analysis Washingtan DC US Gavernment Printing O ice 1994 J Daugman The Importance of Being Random Statistical Principles of Iris Recognition Pattern Recagnitian vol 36 no 2 pp 279291 5 R P Wildes Iris Recognition An Emerging Biometric Technology Prac 0f the IEEE vol 85 no 9 1997 pp 13481363 6 Y 11 T Tan and Y Wang Biometric Personal Identi cation Based on Iris Patternsquot ACTA AUTOMATICA SINICA No1 2002 October 18 2007 WVU 23 Personal Identification BIOM 426 Biometrics Systems Instructor Natalia Schmid August 23 2007 BIOM 426 Outline I Introduction I Applications I Identification methods I Requirements to biometrics I Biometrics technology I Automatic Identification design representation feature extraction matching evaluation I Privacy Issues August 23 2007 BIOM 426 Introduction password August 23 2007 BIOM 426 Identi cation associating identity With an individual Two types of identification problems 0 verification confimiing or enying person39s identity Ami who I claim I am 0 identification or recognition establishing identity Who am I Introduction Facts I Master Card estimated fraud at 450 million per year 1 billion dollars worth of calls are made by cellular bandwidth thieves ATM related fraud 3 billion annually 3000 illegal immigrants crossing the Mexican border each day August 23 2007 BIOM 426 Identi cation methods Person s identity is everything What person represents and believes Engineering approach reduce the problem to i some possession quotsomething What he hasquot or ii some knowledge quotsomething What he knowsquot Another approach reduce it to a problem of authentication based on physical characteristics physiological or behavioral Definition Biometrics are person39s identification based on hisher physiological or behavioral characteristics quotsomething that you arequot August 23 2007 BIOM 426 5 New De nition Biometrics are automated methods of recognizing a person based on a physiological or behavioral characteristic BCC2003 August 23 2007 BIOM 426 H Requirements to biometrics universality everyone should have it uniqueness small probability that two persons are the same in terms of this characteristic permanence invariance with the time collectability can be measured quantitatively performance high identification accuracy acceptability acceptance by people circumvention how easy to fool the system by fraudulent technique August 23 2007 BIOM 426 Accepted Biometrics Accepted and studied biometrics fingerprint voice hand geometry gait ear face iris retina infrared facial and hand vein thermograms key stroke signature DNA DNA signature and fingerprint are recognized in court of law August 23 2007 BIOM 426 8r Biometric Technology Overview Fingerprints are graphical ow like ridges Their formation depends on embryonic development Factors i genetic ii environmental Fingerprint acquisition i scanning inked impression ii life scan Major representations image ridges minutia features derived from ridges or pores Basic approaches to identi cation i correlation based ii global ridge patterns classes iii ridge patterns iv fingerprint minutiae ridge endings and bifurcations v pore pattern August 23 2007 BIOM 426 9 on eof the most acceptable biometrics Two39identi CatiOn approaches i transform eigenvalues analysis of covariance matrix orthonormal basis vectors ii attributebased approach geometric features Factors that in uence recognition i facial disguise ii facial expressions iii lighting conditions iv pose variation August 23 2007 BIOM 426 10 Biometric Technology Overview Iris is one of the most reliable biometrics Frontal images are obtained using near infrared camera 320 X 480 pixels at distance lt 1 meter Iris images are i segmented and ii encoded Twines have different iris patterns August 23 2007 BIOM 426 11 Biometric Technology Overview H WWW Voice print is highly accepted biometrics Used for 39dentification over the telephone Easy to fool the system August 23 2007 Voice is a behavioral characteristic and is not sufficiently unique large database Processing signal subdivided into a few frequency bands The most commonly used feature is cepstral feature log of FT in each band Matching strategies hidden Markoff Vector quantization etc Types of Verification tex edependent texteindependent languageeindependent BIOM 425 12 Biometric Technology Overview Infrared Facial and Hand Vein Thermogram Human bodies radiate heat Infrared sensors acquire an image of heat distribution along the body Images thermograms Imaging methods similar to Visible spectrum photographs Processing raw images are normalized with respect to heat radiating from landmark features In uncontrolled environment other sources of heat could be disturbance August 23 2007 BIOM 426 13 Biometric Technology Overview Gait is the speci c way one walks Complex spatiotemporal behavioral characteristic Gait is not unique and does not stay invariant overtime It is in uenced by distribution of body weight injuries involving joints or brain aging Gait features are derived from a video sequence and consists of characterization of several movements computer vision problem Auguxt23 2007 BIOM 426 14 Biometric Technology Overview Retinal Scan Retinal vasculature is rich in structure Unique characteristic of each individual and each eye Not easy to change or replicate Image capture requires person to peep into an eyepiece and focus on a specific spot A predetermined part of retinal vasculature is imaged Requires cooperation Not accepted by public Can reveal some medical conditions as hypertension August 23 2007 BIOM 426 15 Biometric Technology Overview Signature the way person signs hisher name Highly acceptable behavioral biometrics Evolves over time and depends on physical and mental conditions Easily forged Modeling the invariance and automating signature recognition process is Challenging Two approaches to signature veri cation i static geometric features strokes ii dynamic strokes and acceleration velocity trajectory August 23 2007 BIOM 426 16 Biometric Technology Overview Hand and finger geometry is used for access control 50 of market System captures frontal and side Views of palm Measurements length and width of fingers various distances The representation requirements are only 9 bytes Hand geometry is not unique but highly acceptable August 23 2007 BIOM426 137 Comparison of Biometrics Technologies From Biometrics Personal Identification in Networked Society p 16 August 23 2007 BIOM 426 18 Automatic Identification History Prehistoric Chinese used thumb print for identification Alphonse Bertillon s System of Anthropometric Identification 1882 is based on bodily measurements physical description and photographs 0 Henry s fingerprint classification system 1880 classifies in gt 100 classes Sets of rules are developed for i matching of biometrics ii searching databases Automatic identification is due to inexpensive computer resources advances in computer vision pattern recognition and image understanding August 23 2007 BIOM 426 19 Applications I Civil applications 0 Banking electronic funds transfer ATM security Internet commerce credit card transactions 0 Physical access control airport 0 Information system security access to databases Via login 0 Customs and immigration identification based on hand geometry 0 Voterdriver registration 0 Telecommunications cellular bandwidth access control August 23 2007 BIOM 426 20 Example recognize lhat both he laces in the picture identically sh features and the crown of hair on one of the laces has been digitally manipulated to appear similar to that oi Al Gore 6 August 23 2007 Automatic Identification Design Enrollment 2397 39239 Biometric V 3 Reader Feature Extractor Identification l I Feature Extractor i Biometric E Reader E Feature Matcher j Identification system operates in two modes i enrollment mode and ii identification August 23 2007 BIOM 426 22 Automatic Identification Enrollment mode biometric measurement is captured information from raw data extracted feature person information is stored ID is issued for verification Identification mode biometric is sensed livescan features are extracted from the raw data match is performed search of the database In verification mode person presents ID Then system performs match only against one template in the database August 23 2007 BIOM 426 23 Recognition System Object Acquisition Feature Extraction Training Testing Design Issues Given the speed accuracy and cost specifications 1 How to collect the input data 3D 2D multiple Views high or low resolution 2 Internal representation features for automatic feature extraction 3 How to extract features Algorithms etc 4 How to select the quotmatchingquot metric Measurements are made in specific space 5 How to implement it 6 Organization of database 7 Effective methods for searching a template in the database binning etc August 23 2007 BIOM 426 25 Acquisition Quality of collected data determines perfOrmance of the entire system Associated tasks i quality assessment ii segmentation separation of the data into foreground and background Research efforts i richer data 3D color etc ii metrics for assessment quality of measurements iii realistic models Solutions enhancement August 23 2007 BIOM 426 26 Representation Which machinereadable representation captures the invariant and discriminatory information in the data Determine features st invariant for the same individual intraclass variation maximally distinct for different individuals interclass 0 More distinctive features offer more reliable identification Representation has to be storage space ef cient smart card 2 Kbytes Representation depends on biometrics August 23 2007 BIOM 426 27 Feature extraction Given raw data automatically extracting the given representation is difficult problem Example manual fingerprint system uses about a dozen of features For automatic system many of them are not easy to reliably detect Feature extraction procedures are typically designed in ad hoc manner inefficient when measurements are noisy 0 Determining effective models for features will help to reliably extract them esp in noisy situations August 23 2007 BIOM 426 28 Matching Similarity metric should be robust against noise structural and statistical variations aging and artifacts of feature extraction module Example signature hard to define the ground truth Performance is determined by i representation and ii similarity metric Tradeoff better engineering design vs more complex matcher Example fingerprint variations in features and rigid matcher vs Flexible matcher August 23 2007 BIOM 426 29 Matching Fingerprint Sources of distortion and noise i inconsistent contact 3Dto2D ii nonuniform contact due to dryness of skin sweat dirt humidity in the air etc iii irreproducible contact injuries to the finger iV feature extraction artifacts measurement error V sensing itself adds noise August 23 2007 BIOM 426 30 Evaluation moms August 23 2007 An enduser questions i Does the system makes an accurate identification ii Is the system sufficiently fast iii What is the cost of the system Because of noise distortions and limited infOImation no metric is adequate for reliable identification Decisions genuine individual imp osteri BIOM 426 31 Evaluation Maxehing scare x Au gust 213m quot Genuine distribution pslHInz 1 Four types of outcomes 3 Imposter is rejec 4 Imposteriisf agCQ pg mjm gy m 7613 ail Evaluation Measure of performance ROC receiver operating curve Con dence Intervals Forensic Applican39ons E E 2 r E E Civilian 2 Applications 2 High Security Access 5 Applications False Nonernch Rare man Useful Links httpWWWbiometrics0rg biometric publications and periodicals C O N S O R T I U M research and databases meetings and events httpWWWitlnistg0Vdiv895bi0metrics E MICHIGAN STATE 0 httpblometrlcscsemsuedu u N I v E R s T Y httpWWWWvuedubknc httpWWWciterwvuedu August 23 2007 BIOM 426 34 Iris Recognition BIOM 426 Biometrics Systems Instructor Natalia Schmid October 25 2005 WVU 1 Outline Anatomy Iris Recognition System Image Processing John Daugman iris localization encoding Measure of Performance Results Other Algorithms Pros and Cons Ongoing Work at WVU References October 25 2005 WVU Anatomy of the Human Eye October 25 2005 WVU Eye Camera Cornea bends refracts and focuses light Retina Film for image projection converts image into electtical signals Optical nerve transmits signals to the brain Structure of Iris Iris 2 Aperture Different types of muscles the sphincter muscle constiiction radial muscles dilation Iris is at Color pigment cells called melanin The color texture and patterns are unique October 25 2005 WVU 4 Individuality 0f Iris Left and right eye irises have distinctive pattern October 25 2005 WVU Iris Recognition System Ima e 39 39 AchISlthn g Locallzatlon IriSCOde Gabor Filters Polar Representation 1 Demarcated Zones October 25 2005 WVU 6 Iris Imaging Distance up to 1 meter 0 Nearinfrared camera 0 Mirror October 25 2005 WVU Imaging Systems Walkup Came ras menmm m 0K Rlsp MEWS Fuuy Amnmah cy Two Eye Selzcled Hardware pmnen Ms Reeoem n en Camera 7223 3 2quot my Panasonic L6 E39omnics httpwwwiridiamechcom Fuuy Autumatm Two Eye Ins Recognman Camera Panasumvz amesznn 92mg 2 men may izssu cm Panasnmc BMgtET3DD Hv gh Speed Fixed Focus Twn Eye m Recognition Camera aper39ata a 1245quot pending away an4c cm Aucu Fucu One Eye Iris Recagnicion Camera aDerates ex 3710 away 325 cm 1 LG 3am October 25 2005 WVU 8 Imaging Systems Desktop Cameras httpwww iridiantech com m T the K m Selene sum Mm Panasnnic BMETIDD Rgcggn l n memmmcmg Camera apzrales at 5W0 Software Inc 1921 away 47753 cm Comwmrwulls quot quotquotquot quotquotquot quot Handheld Cameras mmwm OKI IMPASS Tethgred PC m Recnanitinn Camera warm at 1quot 3 pend away 377 cm link Secudmem csNA PER 22 seanamne m tethered ins recugmtiun camera Pendmg Octobet 25 2005 WVU 9 Image Processing John Daugman 1994 Pupil detection circular edge detector max GEO3 We 3r j 1x yds 2717 VJOJ O Segmenting sclera a Y5 2 mun j J39Im WdJdg 7m max i 7 re15ro10rol Br W75 76 r 5 7 October 25 2005 WVU Rubbersheet Model Each pixel Xy is mapped into polar pair r 0 Circular band is divided into 8 subbands of equal thickness for a given angle 0 Subbands are sampled uniformly in 0 and in r Sampling averaging over a patch of pixels October 25 2005 WVU 11 Encoding 2D Gabor filter in polar coordinates 609 zexp 27ia990 r ro 9 90 2 2 a b a 1909 r0 900 601 GABOR FILTER ODD PART GABOR FILTER EVEN PART October 25 2005 WVU IrisCode Formation Setting the Bits in an IrisCode hm lif Bf5imtan e m p4u ewn 391p7wmw 2 0 mm u if 8 j a wt n wk meu er5n 2B JF7 pdpdqg lt a hm 1 if 1ij g iWWr kimiFVu eith 1py pdpd 2 G Am a if1mI eikwf ni ciimi zquot eian ZVp7 3pddeg lt a Intensity is left out of consideration Only sign phase is of importance October 25 2005 WVU 256 bytes 2048 bits Measure of Performance Offline and online modes of operation Hamming distance standard measure for comparison of binary strings 1 n nk1 X and y are two IrisCodes 6 is the notation for exclusive OR XOR Counts bits that disagree October 25 2005 WVU Observations Strong radial dependencies Some angular dependencies October 25 2005 0 Two IrisCodes from the same eye form genuine pair gt genuine Hamming distance Two IrisCodes from two different eyes form imposter pair gt imposter Hamming distance 0 Bits in IrisCodes are correlated both for genuine pair and for imposter pair 0 The correlation between IrisCodes from the same eye is stronger WVU 15 Probabiliiy no 0102 03 04 05 06 07 08 0910 cl Observations Read J Daugman s statement with caution Interpret correctly Independence of Bits across IrisCodes Mk MAM M M w wuww v w Akli lvn axi lMAKWl ywwvw ivwvw October 25 2005 l l u 60 80 100 Code Bii Locaiion 120 The fact that this distribution is uniform indicates that different irises do not systematically share any common structure For example if most irises had a furrow or crypt in the 12 o39clock position then the plot shown here would not be at URL httpwwwclcamacukusersjng000independencehtml WVU Measure of Performance Hamming distance standard measure for comparison of binary strings 1 n D Z xk 6 y k k1 X and y are two IrisCodes G9 is the notation for exclusive OR XOR Counts bits that disagree XOR 169120 Example 100001100011111 G9 169021 101010000011101 0 11 069020 001011100000010 October 25 2005 WVU Training Sets M users 2 iris images per user i M 10000 Genuine Set Imposter Set usermiris1usermiris2 Formed from combination of irises from different users m 1 M userkiris 1 user1iris 1 Compute M genuine Hamming userkiris1userliris2 distances userkiris2userliris1 userkiris2userliris2 k is not equal to 1 k11M Compute imposter Hamming distances October 25 2005 WVU 18 Count 300000 500000 100000 Degrees of Freedom Binomial Distribution of 91 million lrisCode HDs Solid Curve binomial PDF N249 degreesrofrfreedom p0 5 9060003 different ins comparisons mean 0 499 stnd dev 0 0317 min 0334 max 0664 00 01 02 03 04 05 06 07 08 09 10 Hamming Distance October 25 2005 WVU Imposter matching score normalized histogram approximation curve Binomial with 249 degrees of freedom Interpretation Given a large number of imposter pairs The average number of distinctive bits is equal to 249 Histograms 0f Matching Scores Decidability Index d prime Decision Environmentfor Iris Recognition d prime 1136 The cross over point is 0342 different stnd dev n 0197 Compute FMR and FRR for every threshold value Density 25 million comparisons d39 75 Hamming Distance October 25 2005 WVU 20 Decision The same eye distributions depend strongly on the quality of imaging Decision Environment lor Iris Recognition Nonuldeal Imaging Density Non ideal conditions motion blur same different fOCllS mean u EU 10 mean 04555 annuluw mas smuue new p086 varlatlon illumination d 23 rrullion wr rmarlmns it l1 Of 2 03 01 i 7 Elia 09 10 Hamming Distance October 25 2005 WVU 21 Density Decision Ideal conditions Decision Environment for ttis Recognition Meat Imaging same different Ir latttl GJS mudeu p 003 mean 0019 ideJEM wtth d39141 482634 cempansms M 11 02 as m 05 15 0 La 05 Hamming Distance October 25 2005 WVU Imaging quality determines how much the same iris distribution evolves and migrates leftwards dprime for ideal imaging dprime 141 dprime for nonideal imaging previous slide dprime 73 22 Error Probabilities HD Criterion Odds of False Accept Odds of False Reject 028 1 in 1012 1 in 11400 029 1 in 1011 1 in 22700 030 1 in 62 billion 1 in 46000 031 1 in 665 million 1 in 95000 032 1 in 81 million 1 in 201000 033 1 in 111 million 1 in 433000 034 1 in 17 million 1 in 950000 0342 Crossover 1 in 12 million 1 in 12 million 035 1 in 295000 1 in 212 million 036 1 in 57000 1 in 483 million 037 1 in 12300 1 in 113 million Biometrics Personal Identification in Networked Society p 115 October 25 2005 WVU 23 False Accept Rate For large database search FMR is used in verification Binomial Distribution of 91 million IrisCode HDs FAR is used in identification g SOild Curve binomiai PDFY N249 degreesrofrfreed om p0 5 N O 9060003 differentins oom ansons FARzl l FMR zNFMR quot o a mean 0499 stnd dev 0 0817 min 0334 max 0664 Adaptive threshold to keep FAR fixed g HD 2032 0013910g10W 2 agitating 08 09 1 cril October 25 2005 WVU 24 Test Results Wm mwmnmu r Hmur u 9mm Lm USA mm 197quot n mmTmmmLmsl Lnuusm 222743 u Scum mm 1 x mm 49950 M M Euvhul XL mum 199quot n Lywrnm i75 mum mnmm Miuml Plnwlml Lulu L1 301 271 million l Duuguml 11 mm 91 million Ludmn Tw lulnlnglm L39s mm 984 million httpwww clcamacukusersjgdl OOOirlstestspdf October 25 2005 WVU The results of tests published in the period from 1996 to 2003 Be cautious about reading these num ers The middle column shows e number of individuals per dataset Performance Comparison Face Fane2 FPrnhwp FFrcNMZ FPrnphml island 0 ms AIMem mne a r man False Relax Rate m 1 new mm 1quot 1 394 anr n1u mm Amch Ran UK National Physical Laboratory test report 2001 httpWWWclcamacukusersjgd1000NPLsummarygif October 25 2005 WVU 26 Performance Comparison Best0f 3 error rates a Fam Fpmp 0sz AIMem 4km AFFrcmv 2 OFPvnpunal Hana Fa se Rqect Rare aw False Accept Rale UK National Physical Laboratory test report 2001 Odober 25 2005 WVU Other Systems R Wildes et 31 System 1 Image Acquisition 256 pixels across diameter 20 cm distance diffuse source circular polarization and a lowlight level camera 2 Iris Localization image is transformed into a binary edgemap contour tting using Hough transforms 3 Pattern Matching alignment of two patterns representation a Laplacian pyramid goodness of match estimate of correlation coefficient Fisher s linear discriminant Iris Recognition An Emerging Biometric Technology Proc of the IEEE 1997 October 25 2005 WVU 28 Future of Iris October 25 2005 CIA to capture Ins recognmon at a distance Tabassum Zakana Wednesdavy 5 Ndyemher am The u s centran t dence Adency t5 develapmg e able 0 tden fv Usable fmm they are mavmq at a dtstance And w Kyrbyy 52mm thsmal Sme tst at cuts xnt Aqencv was a technmaqy wmch can he HOQOHOUSW maccurate Merencesm mplefactar w dhtmd and cunentmsresadnmanteeh qu expressmn can tmped tdentt u an cf sameane usmg stand tn current tacyax recaqmtmn Qechnabq h satd Ktrbv unthe Thase dtfferences are sd Sign cant that my awn Greenseun Dtcture taken Wu dwerent places at Wu dtfferent ttmes t5 actua mare dtf cultta match than R wauld he a match me wtth sameane m m audtencey he am the farum Ktrbv satd m Dragram whmh was created Wu reqmres a ceraperatyye sumect Wha stand trdnt cf the scanner and hne up the eye prdperxyy he satd t We re ddymd at remdte ms recadn any Kt y satdy 39 addmg that R wauld he mare valuable yr the ms cauld he captured by a camera whde the Dersan was m match at a dtstance a make the tdentmcatmn one cf the mam thrusts cf cur Dragram 5 m factm make ms Dasstble httpwwwabcnatausCiencenewsstoriess982770htm WVU Fraud Protection 1 Hippus steadystate small oscillations of pupil size at about 05 Hz 2 Tracking eyelid movements Nduml Irhl 3 Examining ocular re ections 4 optical surfaces 4 re ections 4 2D Fourier spectra printer s dot in artificial irises El Fuurlur upmlrulll u lululul Ila 2D Fnurlel rpmJurqu ul lulul lrh October 25 2005 WVU 30 References 1 J Daugman s web site URL httpwwwclcamacukusersjgd1000 2 J Daugman High Confidence Visual Recognition of Persons by a Test of Statistical Independence IEEE Trans on Pattern Analysis and Machine Intelligence vol 15 no 11 pp 1148 1161 1993 3 J Daugman United States Patent No 5291560 issued on March 1994 Biometric Personal Identification System Based on Iris Analysis Washington DC U S Government Printing O ice 1994 4 J Daugman The Importance of Being Random Statistical Principles of Iris Recognition Pattern Recognition vol 36 no 2 pp 279291 5 R P Wildes Iris Recognition An Emerging Biometric Technology Proc of the IEEE vol 85 no 9 1997 pp 1348 1363 6 Y Zhu T Tan and Y Wang Biometric Personal Identification Based on Iris Patterns ACTA AUTOMATICA SINICA No1 2002 October 25 2005 WVU 31