自动指纹识别算法研究

烟台大学硕士学位论文自动指纹识别算法研究姓名：公绪成申请学位级别：硕士专业：计算机软件与理论指导教师：孙立民20100301IGaborIIAbstractWiththepropertyofuniquenessanddoesnotchangeallthroughone’slife,fingerprinthasbeenwidelyusedinmanyfields,bothpoliceandciviluse.Withthestrongdemandofautomaticidentifying,fingerprintandotherbiometrictechnologyhavebecomeabigindustryandgrowingquicklyyearly.Theresearchoffingerprintidentifyingalgorithmisalong-termandhighchallengework,manyproblemshavenotbeenresolvedtillnow.Sothissubjectstillneedsintensivestudy.Afingerprintidentifyingalgorithmisdeveloped,includingimagepre-processing,featureextractionandfeaturematching.Themajorcontributionsareasbellows.Firstofall,animprovedpoint-levelorientationfieldestimationmethodisproposed.Thetraditionalblock-levelmethoddoesn’ttoleratestrongnoise,andcan’tdescribetheaccuratedirectioninthesingularityarea.ThenewmethodusingaGaussfiltertoextractthelow-frequencyinformationoforiginalimageandzoomitout,anewsmallimageisobtained.Point-levelorientationfieldiscomputedfromthissmallimage.Usingthismethod,theaccuracyishighlyprovedandtheprocessingspeedisfastenough.Secondly,anewone-dimensionalfilterisdesignedforimageenhancement.Usingthisfilter,thespeedofenhancementishighlyimproved.Theprincipleofexistingenhancementalgorithmislow-passfilteringalongtheridgeandband-passingfilteringverticallytotheridge.Thenewone-dimensionalfilterisdesignedaccordingthisthought,andthecalculationquantityisgreatlyreduced.Finally,inthematchingphase,anewmethodforreferencepointlocationisfirstlyproposed.Apoint-patternmatchingmethodisdesigned.Anewthoughtoftwo-phasematchingincludingfast-modeandaccurate-modeisintroducedandaccuracyisimprovedafterapplyingthismethod.IIIKeywords:fingerprintidentifying,fingerprintsegment,orientationfield,enhancement,featureextraction,featurematching111.1/DNA1a-f1g-i[1](1)(2)(3)(4)[1](1)(2)(3)IT21.1abcdefDNAghi1.1.1311.21.2——Minutiae1KB2430360%40%1/1078[3]41/5055FTEFailToEnroll61.31.36785%1.1.27911IBGInternationalBiometricGroup[4]2008~20141.41.5AFIS/Live-ScanFingerprintFaceRecognitionHandGeometryMiddlewareIrisRecognitionVoiceRecognitionVeinRecognition201493.61.4200981.52009–2014BioAPICompaqIBMIdenticatorMicrosoftMirosNovellIncAPIAPIAPI1.1.391.1Zephyr1.6[4]ZephyrZephyrZephyr1.1101.6Zephyr1.21.2.150%123101041/64011011641.7[5]DNA5ab1.71.2.2123InternetATMPDA1.2.31299.99%0.00001%FVC20064EER5.356%0.169%1.645%0.496%123HenryHenry541.26.16%7.79%17.03%36.48%32.52%131.31Gabor2141.41522.12.12.1FVC2002[6]DB1SETAArchTendedArchLeftLoopRightLoopWhorlTwinLoop[7]2.216PattenArea2.3Core2.32.26Delta2.3RidgeCount172.3Minutiae1150[8]EndingBifurcationDotorIslandEnclosureShortRidgeRidgeBreakingBridgeCrossingSpur2.12183(,)xy42.12.2192.42.52.4ID(1:)NNID/(1:1)ID/ID(1:1)202.51[9]210%[42]34150FBI90%2.22182.62.268.2%22.6%0.7%2.5%3.2%2.112.656VerifyingIdentifying1:11:N22ID2.3FMRFalseMatchRateFNMRFalseNon-MatchRateEEREqual-Error-RateROCReceiveOperatingCurve2.7232.7ROC2.42433.13.110%3.13.2253.23.3(a)(b)(c)(d)(e)(f)3.33.23.3a26[9]020222002((,))((,))(,)((,))()VARIijMMifIijMVARGijVARIijMMotherwiseVAR⎧-+⎪⎪=⎨-⎪-⎪⎩1112001(,)NNijMIijN--===∑∑21122001((,))NNijVARIijMN--===-∑∑3(,)Iij(,)ijMVAR0M0VAR3.3b0M0VAR110703.23.427(a)(b)(c)3.4FFTLinHong[10]:squared-errorclustering62861NNBazen[11,12]XinjianChen[13][14]Gabor[15]Gabor,GaborGaborx,Gabor,Gabor,1x(u,v)y(u,v)G,G(,)(,)xIuvGuvu∂=∂4(,)(,)yIuvGuvv∂=∂5(,)xGuv(,)yGuv(,)Iuv(,)uv3.5Sobel293.5Sobel2x(u,v)y(u,v)G(u,v)=|G,G|G33.41203.33.3.1303.6a[0,180)3.6b[0,180)180°(a)b3.6313.73.3.21.[16]83.73.8,,99,8,3.8i(i=0,1,,78)Gmean[i],84,04,15,26,37,32[]([]-[4]),0,1,2,3GdiffjabsGmeanjGmeanjj=+=,Max=arg(Max(()),iGdiffiMaxiMax+4i,Gray,([])([4])4iMaxifabsGrayGmeaniMaxabsGrayGmeaniMaxiDiriMaxotherwise--+⎧=⎨+⎩6iMax4iMax+2.[10,11,17]WW×500dpiW16(,)ij(,)xGij(,)yGijSobel(,)ij22((,)(,))xxyWVGuvGuv=-∑72(,)(,)yxyWVGuvGuv=∑83311(,)tan()22yxVijVpq-=+.9WW3.[18]3.9BLKSZBLKSZ×3.10(a)WNDSZWNDSZ×OVRLPSTFTq()pq1()sin(2)1{}tan2()cos(2)pdEpdqqqqqqqqq-⎧⎫⎪⎪=⎨⎬⎪⎪⎩⎭∫∫103.934(a)(b)(c)3.103.3.3[6][19,20,21][16,28]0-180180(,)cos(2(,))xijijqΦ=11(,)sin(2(,))yijijqΦ=1235xΦyΦxy22'22(,)(,)(,)wwxxuwvwijWuviuwjvwΦΦΦΦ=-=-Φ=Φ--∑∑1322'22(,)(,)(,)wwyyuwvwijWuviuwjvwΦΦΦΦ=-=-Φ=Φ--∑∑14wf55(,)Wuv'1'(,)1(,)tan()2(,)yxijOijij-Φ=Φ(15)3.113.12[19,20,25]3.11363.12[22][23]5[20,24,25,26]Sherlock[27]Zero-PoleModelVizcaya[28][29]37[20][30]PointCharge4[31][32][33][34]3.3.41233.3.5383.133.131/41/41)1/4Gauss3.14b2)ww×3)(,)xyGGSobel4)θ22((,)(,))xxyWVGuvGuv=-∑(16)2(,)(,)yxyWVGuvGuv=∑(17)11(,)tan()22yxVijVpq-=+(18)5)2)–4)3.14c39a(b)(c)3.143.63.4[10,18][10]1ww×16w=2wl×3.15403.15[26]3l1[]01(,),0,1,...,1,wkdXGuvklw-===-∑(19)()cos(,)()sin(,),22wluidOijkOij=+-+-(20)()cos(,)()cos(,).22wlvjdOijkOij=+-+-(21)40.153.1641[18]3.163.53.5.1[35,36]LinHong[11]Gabor42[37,38]GaborGabor[39][18]rootfiltering[40]Log-Gabor[18]3.5.2P43.0+512RAMFVC2002DB1[6][10]140-200ms[18]100-120ms433.5.3772()exp()cos(0.2)72xGxxp=-223.33.6FVC[10]3.173.17adgjbehk[10]cfil443.171000100403dpi3.183.18[10]1031045100×10=1000100×99=9900ROC3.193.19ROC3.183.13.1Gabor131242128231902139331462229431752437315922333.5463.302552550OTSU[41][42-4