Av26303311
Upcoming SlideShare
Loading in...5
×

Like this? Share it with your network

Share

Av26303311

  • 601 views
Uploaded on

 

  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
No Downloads

Views

Total Views
601
On Slideshare
601
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
14
Comments
0
Likes
1

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311 An Identity-Authentication System Using Fingerprints Mr.Ratnakar anandrao kharade 1, Mr. M.S. Kumbhar H.O.D. 2* 1 Asst.Professor,Department of Electronics and Telecommunication Engg, Dr Daulatrao Aher College Of Engineering,Karad, 2 .Professor,Department of Electronics and Telecommunication Engg, R.I.T. college of engineering, sakharale.Abstract Fingerprint verification is an important The topic of this paper is a verification systembiometric technique for personal identification. based on fingerprints, and the terms verification,In this paper, we describe the design and authentication, and identification are used in aimplementation of a prototype automatic loose sense and synonymously.identity-authentication system that usesfingerprints to authenticate the identity of an Accurate automatic personal identificationindividual. We have developed an improved is becoming more and more important to theminutiae-extraction al-gorithm that is faster and operation of our increas-ingly electronicallymore accurate than our earlier algorithm [58]. interconnected information society [13], [20], [53].An alignment-based minutiae-matching Traditional automatic personal identificationalgorithm has been proposed. This algorithm is technologies to verify the identity of a person, whichcapable of finding the correspondences between use ªsomething that you know,º such as a personalinput minutiae and the stored template without identifica-tion number (PIN), or ªsomething that youresorting to exhaustive search and has the ability have,º such as an identification (ID) card, key, etc.,to compensate adaptively for the nonlinear are no longer considered reliable enough to satisfydeformations and inexact transformations the security requirements of electronic transactions.between an input and a template. To establish an All of these techniques suffer from a commonobjective assessment of our system, both the problem of inability to differentiate between anMichigan State University and the National authorized person and an impostor who fraudulentlyInstitute of Standards and Technology NIST 9 acquires the access privilege of the authorizedfingerprint data bases have been used to estimate person [53]. Biometrics is a technology thatthe performance numbers. The experimental (uniquely) identifies a per-son based on hisresults reveal that our system can achieve a good physiological or behavioral characteristics. It reliesperformance on these data bases. We also have on ªsomething that you areº to make personaldemonstrated that our system satisfies the identification and therefore can inherentlyresponse-time requirement. A complete differentiate be-tween an authorized person and aauthentication procedure, on average, takes fraudulent impostor [13], [20], [53]. Althoughabout 1.4 seconds on a Sun ULTRA 1 biometrics cannot be used to establish an absoluteworkstation (it is expected to run as fast or faster ªyes/noº personal identification like some of theon a 200 HMz Pentium [7]). traditional technologies, it can be used to achieve a ªpositive identificationº with a very high level ofKeywords: Biometrics, dynamicprogramming, confidence, such as an error rate of 0.001%fingerprint iden-tification, matching, minutiae,orientation field, ridge extraction, string matching, A. Overview of Biometricsverification. Theoretically, any human physiological or behavioral characteristic can be used to make aI. INTRODUCTION personal identification as long as it satisfies the There are two types of systems that help following requirements [13]:automatically establish the identity of a person: 1) 1) universality, which means that every personauthentication (verifica-tion) systems and 2) should have the characteristic;identification systems. In a verification system, a 2) uniqueness, which indicates that no twoperson desired to be identified submits an identity persons should be the same in terms of theclaim to the system, usually via a magnetic stripe characteristic;card, login name, smart card, etc., and the system 3) permanence, which means that theeither rejects or accepts the submitted claim of characteristic should be invariant with time;identity (Am I who I claim I am?). In an 4) collectability, which indicates that theidentification system, the system establishes a characteristic can be measured quantitatively.subjects identity (or fails if the subject is not In practice, thereenrolled in the system data base) without the aresomeotherimportantrequirementssubjects having to claim an identity (Who am I?). 303 | P a g e
  • 2. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-3111) performance, which refers to the achievable is a rapidly evolving technology that has beenidentification accuracy, the resource requirements to widely used in forensics, such as criminalachieve an acceptable identification accuracy, and identification and prison security, and has thethe working or environmental factors that affect the potential to be widely adopted in a very broad rangeidentification accuracy; of civilian applications2) acceptability, which indicates to what extent 1) banking security, such as electronic fundpeople are willing to accept the biometric system; transfers, ATM security, check cashing.3) circumvention, which refers to how easy it is tofool the system by fraudulent techniques. Biometricstransactions;2) physical access control, such as applications account for approximately $100airport access control; 3) information system million. The biometric systems intended forsecurity, such as access to data bases via login civilian applications are growing rapidly. Forprivileges; example, by the year 1999, the world market for4) government benefits distribution, such as biometric systemsused for physical access controlwelfare disbursement programs [49]; alone is expected to expand to $100 million [53].5) customs and immigration, such as theImmigration and Naturalization Service PassengerAccelerated Service System (INSPASS) whichpermits faster immigration procedures based onhand geometry6) national ID systems, which provide a unique IDto the citizens and integrate different governmentservices [31];7) voter and driver registration, providingregistration facilities for voters and drivers.Currently, there are mainly nine different biometrictechniques that are either widely used or underinvestigation,including face, fingerprint, handgeometry, hand vein, iris, retinal pattern, signature,voice print, and facial thermograms Although each of these techniques, to acertain extent, satisfies the above requirements andhas been used inpractical systems [13], [18], [20], No metric is sufficiently adequate to give[53] or has the potential to become a valid a reliable and convincing indication of thebiometric technique [53], not many ofthem are identification accuracy of a biometric system. Aacceptable (in a court of law) as indisputable decision made by a biometric system is either aevidence of identity. For example, despite the fact ―genuine individual‖ type of decision or anthatextensive studies have been conducted on ―impostor‖ type of decision, which can beautomatic face recognition and that a number of represented by two statistical distributions, calledface-recognition systemsare available [3], [62], genuine distribution and impostor distribution,[70], it has not yet been proven that 1) face can be respectively. For each type of decision, there areused reliably to establish/verify identity and 2) a two possible decision outcomes, true or false.biometric system that uses only face can achieve an Therefore, there are a total of four possibleacceptable identification accuracy in a practical outcomes: 1) a genuine individual is accepted, 2) aenvironment. Without any other information about genuine individual is rejected, 3) an impostor isthe people in Fig. 1, it will be extremely difficult rejected, and 4) an impostor is accepted. Outcomesfor both a human and a face-recognition system to 1) and 3) are correct, whereas 2) and 4) areconclude that the different faces shown in Fig. 1 incorrect. In principle, we can use the falseare disguised versions of the same person. So far, (impostor) acceptance rate (FAR), the falsethe only legally acceptable, readily automated, and (genuine individual) reject rate (FRR), and themature biometric technique is the automatic equal error rate (EER)2 to indicate thefingerprintidentification technique, which has been identification accuracy of a biometric system [18],used and accepted in forensics since the early [19], [53]. In practice, these performance metrics1970’s [42]. Although signatures also are legally can only be estimated from empirical data, and theacceptable biometrics, they rank a distant second to estimates of the performance are very datafingerprints due to issues involved with accuracy, dependent. Therefore, they are meaningful only forforgery, and behavioral variability. Currently, the a specific data base in a specific test environment.world market for biometric systems is estimated at For example, the performance of a biometricapproximately $112 million. Automatic fingerprint- system claimed by its manufacturer had an FRR ofidentification systems intended mainly for forensic 0.3% and an FAR of 0.1%. An independent test by 304 | P a g e
  • 3. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311the Sandia National Laboratory found that the same practiced fingerprint identificationsystem had an FRR of 25% with an unknown FAR forapproximately 20 years [42]. This discovery[10]. established thefoundation of modern fingerprint identification. In the latenineteenth century, Sir F.B. History of Fingerprints Galton conducted an extensive Fingerprints are graphical flow-like ridges study of fingerprints [42]present on human fingers (see Fig. 2). Theirformations depend on the initial conditions of the C. Design of a Fingerprint-Verification Systemembryonic mesoderm from which they develop. An automatic fingerprint identityHumans have used fingerprints as a means of authentication systemhas four main designidentification for a very long time [42]. Modern components: acquisition, representation(template),fingerprint techniques were initiated in the late feature extraction, and matching.1) Acquisition:sixteenth There are two primary methods of capturing acentury [25], [53]. In 1684, English plant fingerprint image: inked (off-line) and livemorphologist N. scan(ink-less). An inked fingerprint image is typically acquiredin the following way: a trained professional3 obtains animpression of an inked finger on a paper, and the impression is then scanned using a flat-bed document scanner. The livescanfingerprint is a collective term for a fingerprint imagedirectly obtained from the finger without the intermediatestep of getting an impression on a paper. Acquisition ofinked fingerprints is cumbersome; in the context of anidentity-authentication system, it is both infeasible and socially unacceptable for identity verification.4 The mostpopular technology to obtain a live-scan fingerprint image is based on the optical frustrated total internal reflection(FTIR) concept [28]. When a finger is placed on one side of a glass platen (prism), ridges of the finger are in contactwith the platen while the valleys of the finger are not. The rest of the imaging system essentially consists of anassembly of a light emitting diode (LED) light source and a charge-Fig. 2. Fingerprints and a fingerprint classification couple device (CCD) placed on the other side oftheschema of six categories: (a) arch, (b) tented glass platen. The laser light source illuminates thearch, (c) right loop, (d) left loop, (e) whorl, and (f) glass at a certain angle, and the camera is placedtwinloop. Critical points in a fingerprint, called such that it cancapture the laser light reflected fromcore and delta, are marked on (c). the glass. The light that is incident on the plate at Grew published a paper reporting his the glass surface touched bythe ridges is randomlysystematic study onthe ridge, furrow, and pore scattered, while the light incident at the glassstructure in fingerprints, which isbelieved to be the surface corresponding to valleys suffersfirst scientific paper on fingerprints [42].Since totalinternal reflection, resulting in a correspondingthen, a number of researchers have invested a fingerprint image on the imaging plane of thehugeamount of effort in studying fingerprints. In CCD.A number of other live-scan imaging methods1788, a detaileddescription of the anatomical are nowavailable, based on ultrasound total internalformations of fingerprints wasmade by Mayer [16], reflection [61],optical total internal reflection ofin which a number of fingerprintridge edge-lit holograms [21], thermal sensing of thecharacteristics were identified. Starting from 1809, temperature differential (across theridges andT.Bewick began to use his fingerprint as his valleys) [41], sensing of differential capacitancetrademark, whichis believed to be one of the most [47], and noncontact three-dimensional scanningimportant contributions inthe early scientific study [44]. ) Representation (Template): Which machine-of fingerprint identification [42].Purkinje proposed readable representation completely captures thethe first fingerprint classification schemein 1823, invariant anddiscriminatory information in awhich classified fingerprints into nine fingerprint image? This representation issuecategoriesaccording to the ridge configurations constitutes the essence of[42]. H. Fauld, in1880, first scientifically suggested fingerprintverificationdesign and has far-reachingthe individuality anduniqueness of fingerprints. At implications on the design of the rest of the system.the same time, Herschelasserted that he had The unprocessed grayscalevalues of the fingerprint 305 | P a g e
  • 4. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311images are not invariant over the time of representationmight also include one or morecapture.Representations based on the entire gray- global attributes likeorientation of the finger,scale profile ofa fingerprint image are prevalent locations of core or delta,6 andfingerprint class.among the verificationsystems using optical Our representation is minutiae based, and eachmatching [4], [50]. The utility ofthe systems using minutiais described by its location ( coordinates)such representation schemes, however,may be and theorientation. We also store a short segmentlimited due to factors like brightness of the ridgeassociated with each minutia.variations,image-quality variations, scars, and largeglobal distortionspresent in the fingerprint image 3) Feature Extraction: A feature extractor findsbecause these systems areessentially resorting to the ridgeendings and ridge bifurcations from thetemplate-matching strategies for input fingerprint images. If ridges can be perfectlyverification.Further, in many verification located in an input fingerprint image, then minutiaeapplications, terserrepresentations are desirable, extraction is just a trivial task of extracting singularwhich preclude representationsthat involve the points in a thinned ridgemap. In practice, however,entire gray-scale profile fingerprintimages. Some it is not always possible to obtain a perfect ridgesystem designers attempt to circumvent map. The performance of currentlyavailablethisproblem by restricting that the representation is minutiae-extraction algorithms depends heavily onderivedfrom a small (but consistent) part of the the quality of input fingerprint images.finger [50]. If thissame representation is also beingused for identificationapplications, however, then 4) Matching: Given two (test and reference)the resulting systems mightstand a risk of representations,the matching module determinesrestricting the number of unique identitiesthat whether the prints are impressions of the samecould be handled simply because of the fact that finger. The matching phasetypically defines athenumber of distinguishable templates is limited. metric of the similarity between two fingerprintOn theother hand, an image-based representation representations. The matching stage also definesamakes fewerassumptions about the application threshold to decide whether a given pair ofdomain (fingerprints) andtherefore has the potential representationsare of the same finger (mated pair)to be robust to wider varieties offingerprint images. or not.In the case of the minutiae-basedFor instance, it is extremely difficult toextract a representations, the fingerprint-verificationlandmark-based representation from a problem may be reduced to a pointpattern matching(degenerate)finger devoid of any ridge structure. (minutiae pattern matching) problem. InTypically, automatic fingerprint identification and the ideal case, if 1) the correspondence between theauthentication systems rely on representingthe two templateminutiae pattern and input minutiae patternmost prominent structures5: ridge endings and is known,ridge bifurcations. Fig. 3 shows examples of 2) there are no deformations such as translation,ridgeendings and ridge bifurcations. These two rotation,and deformations between them, and 3)structures are background-foreground duals of each each minutia present in a fingerprint image isother, and pressurevariations could convert one exactly localized, thenfingerprint verification istype of structure into the other. Therefore, many only a trivial task of counting the number ofcommon representation schemesdo not distinguish spatially matching pairs between the twobetween ridge endings and bifurcations. Both the images.Determining whether two representations ofstructures are treated equivalently and a finger extracted from its two impressions,arecollectively called minutiae. The simplest of the possibly separated bya long duration of time, areminutiaebased representations constitute a list of indeed representing the same finger is an extremelypoints defined bytheir spatial coordinates with difficult problem. Fig. 4 illustratesthe difficultyrespect to a fixed imagecentric coordinate system. with an example of two images of the same finger.Typically, though, these minimalminutiae-based The difficulty can be attributed to tworepresentations are further enhanced by tagging primaryreasons. First, if the test and referenceeach minutiae (or each combination of minutiae representations are indeed mated pairs, thesubset,e.g., pairs, triplets) with additional features. correspondence between the test andreferenceFor instance, each minutiae could be associated minutiae inthetwo representations isnotwith the orientation of theridge at that minutiae; or known.Second, the imaging system presents aeach pair of the minutiae could beassociated with number of peculiarand challenging situations, somethe ridge count: the number of ridges visitedduring of which are unique to afingerprint image capturethe linear traversal between the two minutiae. scenario.1) Inconsistent contact: the act of sensingTheAmerican National Standards Institute– distorts thefinger. Determined by the pressure andNational Institute ofStandards and Technology contact of the finger on the glass platen, the three-(NIST) standard representationof a fingerprint is dimensional shapeof the finger gets mapped ontobased on minutiae and includes minutiaelocation the two-dimensional surface of the glass platen.and orientation [2]. The minutiae-based Typically, this mappingfunction is uncontrolled 306 | P a g e
  • 5. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311and results in different inconsistently mapped typical finger-imagingsystem distorts the image offingerprint images across theimpressions.2) the object being sensed due to imperfect imagingNonuniform contact: The ridge structure of a conditions. In the FTIRsensing scheme, forfingerwould be completely captured if ridges of the example, there is a geometric distortion because thepart of the finger being imaged are in complete image plane is not parallel tothe glass platen.Inopticalcontact with the glass platen. However, the light of the operational environments mentioneddryness of above,the design of the matching algorithms needs to establish andcharacterize a realistic model of the variations among therepresentations of mated pairs. This model should includethe properties of interest listed below. a) The finger may be placed at different locations on theglass platen, resulting in a (global) translation of the minutiae from the test representation from those inthe reference representation.b) The finger may be placed in different orientations onthe glass platen, resulting in a (global) rotation of theminutiae from the test representation from that of thereference representation.c) The finger may exert a different (average) downwardnormal pressure on the glass platen, resulting in a(global) spatial scaling of the minutiae from the testrepresentation from those in (a) the reference representation.d) The finger may exert a different (average) shear forceon the glass platen, resulting in a (global) sheartransformation (characterized by a shear direction andmagnitude) of the minutiae from the test representationfrom those in the reference representation.e) Spurious minutiae may be present in both the reference and the test representations.f) Genuine minutiae may be absent in the reference or test representations.g) Minutiae may be locally perturbed from their ―true‖location, and the perturbation may be different foreach individual minutiae. (Further, the magnitude ofsuch perturbation is assumed to be small and within a fixed number of pixels.)h) The individual perturbations among the correspondingminutiae could be relatively large (b) (with respectto ridge spacings), but theFig. 4. Two different fingerprint impressions of the perturbations among pairsof the minutiae aresame finger.To know the correspondence between spatially linear.i) Theindividual perturbationsthe minutiae of these two among the correspondingminutiae could befingerprint images, all of the minutiae must be relativelyprecisely localizedand the deformations must berecovered.3) Irreproducible contact: manual work,accidents, etc.inflict injuries to the finger, therebychanging the ridge structure of the finger eitherpermanently orsemipermanently. This mayintroduce additional spurious minutiae.4) Featureextraction artifacts: The featureextractionalgorithm is imperfect and introducesmeasurement errors. Various image-processing Fig. 5. Aligned ridge structures of mated pairs.operations mightintroduce inconsistent biases to Note that the best alignment in one part (top left)ofperturb the location and orientation estimates of the the image results in a large displacements betweenreported minutiaefrom their gray-scale the corresponding minutiae in the otherregionscounterparts. 5) Sensing act: the act of sensing (bottom right).itself adds noise to theimage. For example, residues large (with respect to ridge spacings), butare leftover from theprevious fingerprint capture. A the perturbations among pairs ofthe minutiae are 307 | P a g e
  • 6. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311spatially nonlinear.j) Only a (ridge) connectivity D. An Automatic Identity-Authenticationpreserving transformationcould characterize the SystemWe will introduce a prototype automaticrelationship between the testand reference identity authentication system, which is capable ofrepresentations [73].A matcher may rely on one or automatically authenticatingthe identity of anmore of these assumptions,resulting in a wide individual using fingerprints. Currently, it is mainlyspectrum of behavior. At the oneend of the intended for user authentication.For example, ourspectrum, we have the ―Euclidean‖ matchers,which system can be used to replace passwordallow only rigid transformations among the testand authentication during the log-in session in areference representations. At the other extreme, multiusercomputing environment.wehave a ―topological‖ matcher, which may allowthe mostgeneral transformations, including, say, II. MINUTIAE EXTRACTIONorder reversals.7The choice of assumptions often Fingerprint authentication is based on therepresents verificationperformance tradeoffs. Only matching ofminutiae patterns. A reliable minutiae-a highly constrained systemwith not too demanding extraction algorithm is critical to the performanceaccuracies could get away with restrictive of an automatic identityauthenticationsystem usingassumptions. A number of the matchers in fingerprints. In our system, we have developed atheliterature assume similarity transformation minutiae-extraction algorithm thatis an improved[assumptions a),b), and c)]; they tolerate both version of the technique described in [58].spurious minutiae as wellas missing genuine Experimental results show that this algorithmminutiae. ―Elastic‖ matchers in theliterature performsvery well in operation. The overallaccommodate a small bounded local perturbationof flowchart of this algorithm is depicted in Fig. 7. Itminutiae from their true location but cannot handle mainly consists ofthree components: 1) orientationlargedisplacements of the minutiae from their true field estimation, 2) ridge extraction, and 3)locations [59].Fig. 5 illustrates a typical situation of minutiae extraction and postprocessing.In thealigned ridgestructures of mated pairs. Note that following subsections, we will describe in detailthe best alignment inone part (top left) of the image our minutiae-extraction algorithm. We assume thatmay result in a large amountof displacements theresolution of input fingerprint images is 500between the corresponding minutiae in theother dots per inch.regions (bottom right). In addition, observe thatthedistortion is nonlinear: given distortions at twoarbitrarylocations on the finger, it is not possible topredict thedistortion at all of the intervening pointson the linejoining the two points. In the authors’opinion, a good matcher needs to accommodate notonly global similaritytransformations [assumptionsa), b), and c)] but also shear transformation[assumption d)] and linear [assumption h)] Fig. 7. Flowchart of the minutiae-extraction algorithm.Fig. 6. Architecture of the automatic identity-authentication system. A. Orientation Field Estimation And nonlinear [assumption i)] differential The orientation field of a fingerprintdistortions. Inour experience, assumption j) is too image representsthe intrinsic nature of thegeneral a model to characterize the impressions of a fingerprint image. It plays a veryimportant role infinger, and its inclusioninto the matcher design fingerprint-image analysis. A number ofmethodsmay compromise efficiency and discriminatory have been proposed to estimate the orientationfieldpower of the matcher. In addition, the of fingerprint images [38], [40], [56]. In ourmatchersbased on such assumptions need to use system, anew hierarchical implementation of theconnectivityinformation, which is notoriously algorithm proposedin [56] is used (Fig. 8). Withdifficult to extract fromfingerprint images of poor this algorithm, a fairly smoothorientation-fieldquality. estimate can be obtained. Fig. 9 shows theorientation field of a fingerprint image estimated 308 | P a g e
  • 7. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311with ourhierarchical algorithm.After the orientation ithm on a fingerprint image (512 _ 512) capturedfield of an input fingerprint image is estimated, a with an inkless scanner. (a) Inputsegmentation algorithm, which is based onthe localcertainty level of the orientation field, is used to image. (b) Orientation field superimposed on thelocate the region of interest within the input input image. (c) Fingerprint region. (d) Extractedfingerprintimage. The certainty level of the ridges. (e) Thinned ridge map. (f) Extractedorientation field at pixel is defined as follows: A. minutiae and their orientations superimposed onOrientation Field EstimationThe orientation field the input image.of a fingerprint image represents the intrinsicnature of the fingerprint image. It plays a matching (minutiae matching). The reasonveryimportant role in fingerprint-image analysis. A for this choice is our need to design a robust,number of methods have been proposed to estimate simple, and fast verificationthe orientationfield of fingerprint images [38], [40],[56]. In our system, a new hierarchical algorithm and to keep a small templateimplementation of the algorithm proposedin [56] is size. A number of point pattern matchingused (Fig. 8). With this algorithm, a fairly algorithms have been proposed in the literature [1],smoothorientation-field estimate can be obtained. [55], [63], [66], [69], [71]. A general pointFig. 9 shows theorientation field of a fingerprint matching problem is essentially intractable.image estimated with ourhierarchical algorithm. Features associated with points and their spatial properties, such as the relative distances betweenIII. MINUTIAE MATCHING points, are widely used in these algorithms to Fingerprint matching has been approached reduce the exponential number of search paths.from several different strategies, like image-based[4], [50] and ridgepattern matching of fingerprint A. Alignment of Point Patternsrepresentations. There also exist graph-based Ideally, two sets of planar point patternsschemes [22], [23], [26], [27], [34], [36] for can be aligned completely by only twofingerprint matching. Our automatic corresponding point pairs. A true alignmentfingerprintverification algorithm instead is based between two point patterns can be obtained byon point pattern testing all possible corresponding point pairs and selecting the optimal one. Due to the presence of noise and deformations, however, the input minutiae cannot always be aligned exactly with respect to those of the templates. To accurately recover pose transformations between two point patterns, a relatively large number of corresponding point pairs need to be used. This leads to a prohibitively large number of possible correspondences to be tested. Therefore, an(a) (b) alignment by corresponding point pairs is not practical even though it is feasible. B. Aligned Point Pattern Matching If two identical point patterns are exactly aligned with each other, then each pair of corresponding points are completely coincident. In such a case, point pattern matching can be simply achieved by counting the number of overlapping pairs. In practice, however, such a situation is not (c) (d) encountered. On the one hand, the error in determining and localizing minutiae hinders the alignment algorithm to recover the relative pose transformation exactly, while on the other hand, our alignment scheme described in Fig. 13 does not model the nonlinear deformation of fingerprints, which is an inherent property of fingerprint impressions.(e) (f) 309 | P a g e
  • 8. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311IV. EXPERIMENTAL RESULTS incorrect otherwise. A total of 489 300 (700x 699)A. Feature-Extraction Performance matchings were performed. It is very difficult to assess theperformance of featureextraction algorithms D. Authentication Testindependently. Accuracy of the extracted minutiae In test 1, for each individual, we randomlywas subjectively confirmed in two ways. Visual selected three fingerprint images that passed theinspection of a large number of typical quality check as the template minutiae patterns forminutiaeextraction results showed that our the individual and inserted them into the systemalgorithm rarely missed minutiae in fingerprint data base. The major reason why weuse threeimages of reasonable quality. fingerprint templates is that a significant number of acquired fingerprint images from the same finger inB. System Performance the MSU data base do not have a sufficient amount We have tested our system on the MSU of common region of interest due to thefingerprint data base. It contains ten images (640 unrestricted acquisition process.480) per finger from 70 individuals for a total of700 fingerprint images, which were captured with a V. SUMMARY AND CONCLUSIONSscanner manufactured by Digital Biometrics. When We have introduced an automatic identity-these fingerprint images were captured, no authentication system using fingerprints. Therestrictions on the position and orientation of implemented minutiaeextraction algorithm is muchfingers were imposed. more accurate and faster than our previous feature- extraction algorithm [58]. The proposed alignment- based elastic matching algorithm is capable of finding the correspondences between minutiae without resorting to an exhaustive search. The system types of inputs. It should not be confused with a practical system. In practice, a number of mechanisms need to be developed besides the minutiae extraction and minutiae matching. A biometric system based solely on a single(a) (b) biometric feature may not be able to meet the practical performance requirement in all aspects. By integrating two or more biometric features, overall verification performance may be improved. For example, it is well known that fingerprint verification tends to have a larger false reject rate due to the reasons discussed above, but it has a very low false accept rate. On the other hand, face recognition is not reliable in establishing the true identity but it is efficient in searching a large data base to find the top matches. By combining(c) fingerprint matching and face recognition, the false (d) reject rate may be reduced without sacrificing the false accept rate, and the system may then be able to operate in the identification mode. Currently, weFig. 15. Results of applying the matching algorithm are investigating a decision-fusion schema toto an input minutiae set and a template. (a) Input integrate fingerprint and face.minutiae set. (b) Template minutiae set. (c) The expected error rate of a deployedAlignment result based on the minutiae marked biometric system is usually a very small number (with green circles. (d) Matching result where 1%). To estimate such a small number reliably andtemplate minutiae and their correspondences are accurately, large representative data sets that satisfyconnected by green lines. the two requirements mentioned in Section I are needed. Generally, under the assumption ofC. Matching Scores statistical independence, the number of tests We first evaluated the matching scores of conducted should be larger than ten divided by thecorrect and incorrect matches. In test 1, each error rate [24]. Currently, we are evaluating thefingerprint in the MSU fingerprint data base was system on a large data set of live-scan fingerprintmatched with all the other fingerprints in the data images.base. A matching was labeled correct if thematched fingerprint was from the same finger, and 310 | P a g e
  • 9. Mr. Ratnakar anandrao kharade, Mr. M.S. Kumbhar / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 6, November- December 2012, pp.303-311REFERENCES challenges and public policy issues,‖ Info. [1] N. Ansari, M. H. Chen, and E. S. H. Hou, Technol.People, vol. 7, no. 4, pp. 6–37, ―A genetic algorithm for point pattern 1994. matching,‖ in Dynamic, Genetic, and [14] L. Coetzee and E. C. Botha, ―Fingerprint Chaotic Programming, B. Souˆcek and the recognition in low quality images,‖ IRIS Group, Eds. New York: Wiley, 1992, Pattern Recognit., vol. 26, no. 10, pp. ch. 13. 1441–1460, 1993. [2] ―American national standard for [15] T. H. Cormen, C. E. Leiserson, and R. L. information systems—Data format for the Rivest, Introduction to Algorithms. New interchange of fingerprint information,‖ York: McGraw-Hill, 1990. American National Standards Institute, New York, NY, Doc. No.ANSI/NIST- CSL 1-1993. [3] J. Atick, P. Griffin, and A. Redlich, ―Statistical approach to shape from shading: Reconstruction of 3D face surfaces from single 2D images,‖ Neural Computation, to be published. [4] R. Bahuguna, ―Fingerprint verification using hologram matched filterings,‖ in Proc. Biometric Consortium Eighth Meeting, San Jose, CA, June 1996. [5] H. Baird, Model Based Image Matching Using Location. Cambridge, MA: MIT Press, 1984. [6] D. H. Ballard, ―Generalized Hough transform to detect arbitrary patterns,‖ IEEE Trans. Pattern Anal. Machine Intell., vol. PAMI-3, no. 2, pp. 111–122, 1981. [7] MFLOPS Benchmark Results. (Feb. 1997.) [Online]. Available: ftp://ftp.nosc.mil/pub/aburto/_flops_1.tbl. [8] T. Biggs, personal communication, Department of Immigration and Naturalization Services, 1997. [9] F. Bouchier, J. S. Ahrens, and G. Wells. (1996). Laboratory evaluation of the iriscan prototype biometric identifier. [Online].Available:http://infoserve.library. sandia.gov/sand_doc/1996/ 961033.pdf. [10] J. P. Campbell, Jr., L. A. Alyea, and J. S. Dunn. (1996). Biometric security: Government applications and operations. [Online]. Available: http://www.vitro.bloomington.in.us:8080/ BC/. [11] J. Canny, ―A computational approach to edge detection,‖ IEEE Trans. Pattern Anal. Machine Intell., vol. PAMI-8, no. 6, pp. 679–698, 1986. [12] G. T. Candela, P. J. Grother, C. I. Watson, R. A. Wilkinson, and C. L. Wilson, ―PCASYS: A pattern-level classification automation system for fingerprints,‖ National Institute of Standards and Technology, Gaithersburg, MD, NIST Tech. Rep. NISTIR 5647, Aug. 1995. [13] R. Clarke, ―Human identification in information systems: Management 311 | P a g e