Fingerprint Recognitio n Technology: Liveness Detection, Image Quality and Skin Disepses Martin Drahansky Brne University of Tecnnology, Faculty of Information Technology Bozetnchova 2, 612 66 Bcno, Czech Republic http://www.fit.vutbr.cz/~drahan | strcncoe.fit.vutbr.cz drahan@fit.vutbr.cz BRNO UNIVERSITY OF TECHNOLOGY FACULTY OF INFORMATION TECHNOLOGY STRaDe © Presentation with copyright March 28, 2011 Introduction General biometric system Data capture Data storage Matching Decision Introduction I 3/42 Skin structure on fingers Introduction I 4/42 Technologies of fingerprint sensors Introduction | 5/42 Other problems at sensorics Dactyloscopic card IIP! ill mix. Earth dust Metallic dust Fine sand Oiled finger About -10°C About +50°C Introduction I 6/42 Security of biometric systems Introduction | 7/42 Finger fakes I m Fake fingerprints of different materials Introduction | 8/42 How to produce a fake finger(print)? Introduction | 9/42 Liveness detection Perspiration Clarkson/Virginia universities 600 sweat glands in 1 inch2 Duration approx. 5 seconds High intra-class variability Liveness detection | 11/42 Spectroscopic characteristics UN • Lumidigm Ltd.(Albuquerque) • Clones: TST Biometrics GmbH, Sagem Morpho etc. 650 700 750 800 850 900 950 300 400 500 600 700 Wavelength (nm) Wavelength (nm) Liveness detection | 12/42 Ultrasonic technology Ultra-Scan / Optel / y v yi Sound wave pulse transmission >^ Sound wave echos, captured to produce images: Echo #1 Echo #2 Echo #3 n T—1—r 1'' Desired image depth is selected by range gate Platen Ridge structure Air gap or contamination Liveness detection | 13/42 Temperature + Hot & cold stimulus • Temperature 17/02/05 16.42.17_|e=0.97 |_| 17/02/05 15.52.43_|e=0.97 |_| 17/02/05 1G.22.20_|e=0.97 | • Hot and cold stimulus Liveness detection | 14/42 Pressure stimulus National utility model ÜPV 19364 Ridge elasticity • 20% Change of color (RGB) G ~ 42 Liveness detection | 15/42 Electrical properties Bio-impedance Resistance / Conductivity Measurement with DC low voltage R_Little R_Ring R_Middle R_Index R_Thumb L_Little L_Ring L_Middle L_Index L_Thumb r D f- --A -A f- -—t\ t- -—A r 0 f- f- --i t- -ii r i i i 0 200 400 600 800 1000 Liveness detection | 16/42 Pulse I. • Heart activity Normal Systolic Dysfunction The ventricles pump The ventricles pump out aboul 60% of oul less than 40 to 50% (he blood, of the blood, Diastolic Dysfundion out about 60% ol the blood, but Itie amount may be lower than normal- Liveness detection | 17/42 Pulse II. • International patent WO/2007/036370 High resolution Common optical camera with fingerprint scanner macro-objective Liveness detection | 18/42 Pulse III. LJveness detection | 19/42 Pulse IV. Macroobjective II. 14 12 10 23 peaks in 15 sec correspond to 92 heart beats in 1 min Real heart beats: 76 / min ir w w w ir ir if 1 7 13 19 25 43 49 55 61 67 73 79 85 91 97 109 115 121 127 133 139 F75l 81 193199 -2 -4 Image number 8 6 4 0 Liveness detection | 20/42 Pulse V. Laser I. CCD camera Liveness detection | 21/42 Pulse VI. I m Liveness detection | 22/42 Blood oxygenation • Infrared illumination (660 nm / 940 nm) • Reflection vs. transmission Finger veins Liveness detection I 23/42 Image quality Fingerprint recognition process Flowchart results of the minutiae extraction Input Image Orientation Field Extracted Ridges Thinned Ridges Minutiae Points Change of image quality Mm Contra Uncompressed JPEG WSQ change 3 \ All ^ mlP mm tyy&'X*' Missing papillary lines Image quality | 25/42 Used sensors for tests of image quality • Supreme Evaluation Development Kit SFM3xxx Model / Features SFM3000 SFM3010 SFM3020 SFM3050 Sensor FingerLoc AF-S2 by AuthenTec FingerChip by Atmel® not known TouchChip® TCS2 by UPEK Technology [Dra25] e-field thermal, sweep optical capacitive Power supply 3.3 V (DC) 3.3 V (DC) 3.3 V(DC) 3.3 V(DC) Take-off current 100 -300 mA 4.5 mA not known not known Resolution [DPI] 250 500 500 500 Sensor size [nun] 13 x 13 11.6 x 0.4 16 x 19 10.4 x 14.4 Module size [WxDxH] [nun] 55 x 40 x 8 55 x 40 x 8 55 x 40 x 8 55 x 40 x 8 Image size [pix.] 128 x 128 360 x 500 272 x 320 256 x 360 Image quality | 26/42 Fingerprint Image Contrast & Histogram • Michelson contrast = v max min / • Lmax-intensity of foreground (papillary line/ridges) • Lmin - intensity of background (valleys) • Weber contrast cWeber =DL L • DL - intensity difference between foreground and background; L - intensity of background Hist(rk) = —, k = 0,1,2,....L — 1 • Fingerprint image histogram g_ , . 1600 • rk - /cth gray grade value 1200 • nk - number of pixels in channel rk : • L - number of gray grades I 200 • n - sum of pixels in the image Gray values Image quality | 27/42 Determination of contrast ratios Used Michelson contrast Higher values (close to 1) are better results (SFM3000) CM=0.98441 CM=0.26393 «3 o o o 0) 1,2 0,8 0,6 0,4 0,2 0 SFM3000 SFM3010 SFM3020 SFM3050 1 70 139 208 277 346 415 484 553 622 691 760 829 898 967 1036 1105 1174 Sequence number of the fingerprint Image quality | 28/42 Histogram normalization & Mean value Histogram normalization • Probability distribution function in range <0;1> Mean value M-1 255 SL = ^ hn (i) SR = ^ hn (i) i=0 i =M • M - mean value Ideal case: M = 128 [Sl=SR] • Theoretical M: mt = dark l'sl" = 2 Bx - start & end of histogram Deviation D: d = M • 100% Q Q_ "O (D N 0 8 05 0.4 0.3 0 2 0.1 o 50 100 150 200 250 300 Gray values w w I 50 100 150 200 250 300 Gray values Image quality | 29/42 Number of papillary lines • In dactyloscopic literature defined as number of papillary lines between delta and core • For estimation of fingerprint's center could be used horizontal and vertical values of papillary lines (comparison with homocentric circles) Image quality | 31/42 Number of papillary lines I m Sensor SFM3000 SFM3010 SFM3020 SFM3050 Horizontal minimum 6.00 6.00 9.00 10.00 Horizontal average 12.86 19.30 19.93 21.25 Horizontal maximum 23.00 27.00 30.00 31.00 Vertical minimum 5.00 3.00 11.00 11.00 Vertical average 13.26 33.18 22.79 25.67 Vertical maximum 24.00 51.00 31.00 37.00 Horizontal Horizontal average Vertical Vertical average .£ 25 j5 20 q. 15 20 15 -10 5 0 !'.'. „1 . r„ ,J.,M, . w h\ \ J'laV \< v,,,',.f, *, u iA«.. ,V SFM3000, stable 1 40 79 118 157 196 235 274 313 352 391 430 469 508 547 586 625 664 703 742 781 820 859 898 937 976 1015 1054 1093 1132 1171 Sequence number of the fingerprint 30 SFM3010, unstable Horizontal Horizontal average Vertical Vertical average ififfl 1 fin fik'K'lA.illifl ( haul .hi k Juii.b . IllHM* 1T,pn||l||j'll I Hr u 80 190 200 218 220 23& 240 250 260 270 280 290 300 310 320 — yirffw^fMflww www v iff I f If v IF i w w^ w pixels -» • Application of sine function ti the crosscut Image quality | 33/42 Sinusoidal shape II. Deviation of the papillary line curvature from the sine function • 100% V Asin J • where afp = j f (x)dx , Asin = Jsin(x)dx , Xs = -p/2, XE = 3p/2 XS XS Deviation of thickness of papillary line from normalized state Th V 0.033 _ 1 100% J 2.54 NPiX [cm] • where Th r numlD^r of pixels resolution of the sensor Image quality | 34/42 Sinusoidal shape III. Deviation of steepness of papillary line from a normalized state a- 60° Deviation of the upward angle D«= 60 Deviation of the downward angle dp=bb-60 • 100% 60° • 100% where f a = arcsm A/Px1 + p f ß = arcsin y J X2 V VPx2 + Py y J 150 152 158 180 182 164 Image quality | 35/42 Filtering in spatial domain Normalization - reduction of variations in gray-level values along ridges and valleys Orientation image estimation - orientation tendency of papillary lines in local neighborhood Frequency images estimation - frequency of ridge and valley structures in local neighborh. along orient. Region mask generation - differentiation of pixels to unrecoverable (non-ridge) and recoverable (ridge) Normalization Enhanced image Orientation image estimation Frequency images estimation Region mask generation Input image Filtering Image quality | 36/42 Image quality | 37/42 Filtering in frequency domain • FFT Filtering IFFT • Used filters for fingerprint enhancement • Butterworth filter • Maximally flat magnitude filter • Ikonomopoulos filter • Based on Lin & Hong approach • Low-pass filter • Classical filtering, however have good results • Chebyshev filter • Comparable with Butterworth filter, different frequency response Image quality | 38/42 Skin Diseases Classification of skin diseases • Change of papillary line structure • Change of skin color • Change of pap. line structure and skin color Skin diseases | 40/42 41/42 Laboratory of biometric systems I m 42/42 Thank you for your attention.